Air pollutant concentrations at more than 1000 national ambient air quality monitoring sites are released by the Ministry of Environmental Protection (MEP) of China and can be downloaded from the China National Environmental Monitoring Center (CNEMC, http://www.cnemc.cn/sssj/, last access: 13 April 2021). The technical requirements for the monitoring system including the composition, installation, operation, maintenance and data quality control are addressed in the China Environmental Protection Standards “HJ 193-2013” (http://www.mee.gov.cn/ywgz/fgbz/bz/bzwb/jcffbz/201308/W020130802493970989627.pdf, last access: 13 April 2021) and “HJ 655-2013” (http://www.mee.gov.cn/ywgz/fgbz/bz/bzwb/jcffbz/201308/W020130802492823718666.pdf, last access: 13 April 2021). In this paper, the hourly concentrations of SO2, NO2, O3, CO, PM2.5 and PM10 at 1412 monitoring sites in 2014 were utilised from CNEMC. The locations of the monitoring sites are shown in Fig. 2. The observational data were used to evaluate the simulated air pollution over China.

The first version of the United Kingdom Earth System Model (UKSEM1) is jointly developed by Natural Environment Research Council (NERC) and the Met Office Hadley Centre and was released in February 2019 (Sellar et al., 2019). UKESM1 is based on the physical climate model HadGEM3 (Hewitt et al., 2011; Kuhlbrodt et al., 2018; Williams et al., 2018) and couples additional components, including a land biogeochemistry model (JULES; Clark et al., 2011), the UK Chemistry and Aerosols model (UKCA; Archibald et al., 2020; Mulcahy et al., 2018), a dynamic vegetation model (TRIFFID; Cox, 2001) and an interactive ocean biogeochemistry model (MEDUSA; Yool et al., 2013). In this study, we used its atmosphere-only (UKESM1-AMIP) version to study the radiative effect of aerosols on ozone. Unlike the fully coupled UKESM1, the atmosphere-only configuration does not include ocean and sea ice models (NEMO/CICE), MEDUSA, or TRIFFID. Instead, UKESM1-AMIP uses prescribed, observation-based sea surface temperatures and sea ice data (https://pcmdi.llnl.gov/mips/amip/, last access: 13 April 2021). The model input for vegetation and surface ocean biology fields is provided by the UKESM1 historical simulations for the Coupled Model Intercomparison Project Phase 6 (CMIP6).

The core atmospheric model of UKESM1-AMIP is the 11.1 version of the Met Office Unified Model (UM; Walters et al., 2019), in which the atmospheric chemistry and aerosols are modelled by UKCA. The new Global Model of Aerosol Processes (GLOMAP-mode; Mann et al., 2010) is a size-resolved aerosol microphysics model. It is used for aerosol simulation in UKCA, including the mass and number of sulfate, black carbon, organic carbon and sea salt. Dust aerosols are not available yet in GLOMAP-mode, and so a bin scheme for mineral dust (Woodward, 2001) is used. The photolysis scheme in UKCA is Fast-JX (Telford et al., 2013), which provides the full scattering calculation for 18 wavelength bins over 177–850 nm. Fast-JX allows the calculation of the interactive photolysis rates in the troposphere (Wild et al., 2000) and improves the calculation of photolysis rates in the stratosphere (Bian and Prather, 2002). In order to focus on ADE-POL and ADE-MOI effects (see Sect. 1), Fast-JX has not been coupled with the GLOMAP-mode aerosol scheme, which means that photolysis rates are independent of the aerosol loading (Sellar et al., 2019).

Two sensitivity simulations were performed to evaluate the radiative effects of aerosols on ozone: (1) EXPradon, which includes the aerosol direct radiative feedback on atmospheric chemistry, and (2) EXPradoff, which is without this radiative feedback. The simulation covers 1 year, i.e. from 1 January 2014 to 31 December 2014. The atmospheric horizontal resolution of UKESM1-AMIP is N96 (∼ 140 km) and the vertical resolution is 85 levels. Emissions are the year 2014 CMIP6 emissions for all runs.

The model performance was evaluated by comparing the simulation results at the surface layer with the ground-based observations. The simulation with radiation feedback (EXPradon) was carried out as the control test. Figure 3 shows annual average concentrations of O3, CO, NO2, SO2, PM2.5 and PM10 simulated in EXPradon along with the concentrations observed at monitoring sites. Pearson's correlation coefficient (R) and mean bias error (MB) are shown in Table 1, using daily average concentration data. In terms of the spatial distribution, the simulation results are found to be in fairly good agreement with the observations. With the economic and industrial development in north and east China, anthropogenic emissions lead to increased air pollution in these areas (M. Li et al., 2017). The model captures the high SO2, CO and NO2 concentrations and the high aerosol loading in north and east China. However, the model produces much higher SO2 concentrations than the observations, most likely due to an overestimation of the emissions. Under the clean air policies, the SO2 emission has declined by 62 % during 2010–2017 (Zheng et al., 2018), but the CMIP6 emissions do not capture this reduction, with 2014 SO2 emissions being higher by 48 % when compared to the region-specific Multi-resolution Emission Inventory for China (MEIC) (Paulot et al., 2018). For the spatial distribution of ozone, the model is in good agreement with observations. The simulated ozone concentration is well correlated with the observed values, with R reaching a maximum of 0.8 in the Jing–Jin–Ji (JJJ) area. The radiation effect improved the model performance in China. When including the meteorological feedback of radiation effect, the average MB of ozone dropped from 10.03 to 5.63, while the average R remained the same (Table 1). In most areas, the correlation between observed and simulated CO, NO, SO2 and particulate matter were higher in EXPradon than in EXPradoff, indicating that including these effects improves the simulation of tropospheric pollutants. Subsequently, we examine these effects in more detail.

The aerosol effect on meteorology was assessed by taking averages over the 1-year simulation and taking the difference between EXPradon and EXPradoff. Figure 4 shows the changes in net downwelling surface shortwave radiation, turbulent kinetic energy, planetary boundary layer height (PBLH) and 10 m wind due to the direct effect of aerosols on radiation. Shortwave radiation is generally reduced due to aerosols over China, and the largest reduction is found in more aerosol-rich parts of the country (Fig. 3l and m), i.e. north and east China. Shortwave radiation was reduced by 30.24 Wm-2 (18.85 %), 19.73 Wm-2 (12.98 %), 20.45 Wm-2 (11.22 %) and 16.27 Wm-2 (13.53 %) in JJJ, the Yangtze River Delta (YRD), the Pearl River Delta (PRD) and the Sichuan Basin (SCB), respectively (Fig. 4a). The high-resolution regionally focused Weather Research and Forecasting model coupled with Chemistry (WRF-Chem) simulation performed by Wang et al. (2016) similarly showed that due to ADE, the solar radiation in China decreased by 20 Wm-2 and the percentage changes ranged from 11.7 % to 14.3 % in different areas. A decreased downwelling solar radiation could cool the surface and cause weaker thermal turbulence in the boundary layer (Liu et al., 2018; Quan et al., 2013). The temperature at 1.5 m is found to be reduced in the North China Plain and south-west China (Fig. S1a in the Supplement) due to the radiation changes. Turbulent kinetic energy (TKE) (Fig. 4b) showed the largest change in north China (JJJ), with a decline of 0.12 m2s-2 (33.43 %), which is consistent with China's largest SW radiation change area. This is in line with the findings of Wang et al. (2020), who found that during a haze episode in winter, the TKE in Beijing declined by 0.1–0.7 m2s-2 due to the aerosol-induced effect. The reduction in TKE in the YRD reaches 23.09 % in our findings, which is the second-highest TKE reduction region in China.

The growth of the boundary layer mainly depends on the atmospheric thermal structure and turbulent exchange intensity (Garratt, 1994; Serafin et al., 2018). Owing to the reduced solar radiation and TKE, the development of the PBL was suppressed and resulted in a shallower and more stable boundary layer (Fig. 4c and d). In north China (JJJ), the annual average planetary boundary layer height (PBLH) was reduced by 102.28 m (22.01 %) due to the ADE. Observations in this area also showed that the average PBLH was reduced by 334–710 m during severe pollution periods compared to clean days (Tang et al., 2016; Zhang et al., 2015). The annual delineation of PBLH in the YRD was 53.39 m (16.26 %), and this reduction was consistent with the WRF-Chem simulation by Wang et al. (2016), who found that the PBLH in east China decreased by 75.2 m in spring, while in other regions of China it decreased by 75–138 m. Using the Weather Research and Forecasting and Community Multiscale Air Quality (WRF-CMAQ) model, Nguyen et al. (2019) also found that the ADE could reduce the annual average PBLH in East Asia by 46.47 m (8.13 %). The lower boundary layer caused by aerosols is usually also accompanied by calm winds and higher relative humidity values (Yin et al., 2019). Here, the 10 m wind is found to be lowered by 1 % to 7.5 % (Fig. 4d) and relative humidity at 1.5 m increased with a maximum of 5.7 % (Fig. S1b). The variations in wind and boundary layer stability would influence horizontal transport and pollutants and moisture accumulation, as well as the vertical dispersion and the exchange of clean air with the free troposphere.

The aerosol direct radiative feedback was found to reduce solar radiation, which resulted in the suppression of PBL height and turbulent intensity, while the suppressed PBL in turn limits the export and diffusion of pollutants. Figure 5 shows the influence of ADE on surface CO, SO2, NO, NO2, PM2.5 and PM10 concentrations. Overall, pollutant concentrations increased when including aerosols, due to the decreasing wind speeds and PBLH. The CO increase caused by ADE averaged over China was 11.04 %, with the biggest changes appearing in north China (JJJ), east China (YRD) and central China (up to 12.25 %–16.17 %). The distribution of SO2, NO and NO2 changes is similar to that of CO, with increases of 5.66 %–38.99 %, 7.71 %–55 % and 2.78 %–40.63 %, respectively. For fine and coarse aerosols (PM2.5 and PM10), the increases are between 9.5 % and 18.6 % in the four selected areas and the spatial distributions of changes are similar to those of gaseous pollutants. Changes in gas and aerosol pollutants were the result of the changes in meteorological conditions. The shallower PBLH reduced the vertical dispersion and compressed the pollutants in the PBL, resulting in higher surface pollutant concentrations. The increased boundary layer stability and reduced wind speed also led to the accumulation of pollutants at their emission sources. The spatial distribution of the changes in pollutant concentration is similar to the spatial distribution of meteorological condition changes and emission sources. With a larger population and more developed industries, north and east China were considered to be the high-emission areas of the country (Wang, 2015; Zheng et al., 2018). These areas are more sensitive to the accumulation of pollutants and showed a stronger increase in the pollutant concentrations due to aerosol effects. Western China is less developed than the eastern parts, and its population and anthropogenic emissions are also lower (Saikawa et al., 2017; Shi et al., 2014). As a result of this, the ADE in west China caused a small increase and even a decrease in pollutant concentrations. In south-west China, the SCB is more developed than the surrounding cities, and its bowl-shaped topography helps trap air pollutants (Ning et al., 2018). More pronounced increases in pollutants' concentrations are also found in this area, but the magnitude is lower than that in JJJ and the YRD. Changes in air pollutants (including NO and NO2) in different regions affect the ratio of NO/NO2, which is related to the loss and the production process of ozone. The change in NO/NO2 and ozone will be further discussed in Sect. 3.5.

The changes in boundary layer stability and PBLH would not only contribute to the pollutant accumulation but would also be linked to the moisture accumulation. The change in horizontal water vapour flux over the land area is small (Fig. S2 in the Supplement). However, a low PBLH could limit the vertical transport of water vapour from the boundary layer to the free troposphere. Figure 6 shows the vertical profile of changes in specific humidity in different parts of China. In most seasons, climatological specific humidity increases in the lower troposphere and drops in the higher layers. In JJJ, the area most affected by ADE, the surface moisture content increases more when comparing EXPradon with EXPradoff, with a maximum change of 4.28 % (6.55 × 10-4 kgkg-1) in June. The annual mean specific humidity decreased by a maximum of 1.69 % (1 × 10-4 kgkg-1) at 1493 m in China.

When more water vapour was trapped in the lower troposphere, there would be less moisture to form cloud in the upper layers (Allen et al., 2019). The annual average cloud amount decreases by 4 % due to aerosol effects on radiation over the whole country (Fig. 7). The area with the largest decline is the YRD with a percentage of 5 %. The cloud optical depth also drops by 7 %–15.6 % in China, with the regional distribution of changes being similar to the cloud amount changes. Clouds attenuate solar radiation, leading to diminished photolysis rates beneath the cloud (Tang et al., 2003; Voulgarakis et al., 2009a, b, 2010). Therefore, the increased water vapour in the PBL results in higher photolysis rates by reducing clouds. However, the increased water vapour in the PBL will also enhance extinction by aerosol hygroscopic growth, which results in lower photolysis rates. Figure 7 shows that surface photolysis rates JNO2 and JO1D both increase, which means, comparing them to the aerosol hygroscopic growth, that the aforementioned cloud reduction is the dominant effect. The national average JNO2 and JO1D rose by 4.1 % and 3.3 %, respectively. The SCB is the region with the largest increase in JNO2 and JO1D, with percentage increases of 8 % and 7.9 %, respectively. The increase in JNO2 and JO1D could lead to an increase or decrease in ozone concentration.

The meteorological feedbacks that we study, ADE-POL and ADE-MOI, may have contrasting effects on ozone. For ADE-POL, the relationship between NO and NO2 concentrations could be used to predict the changes in ozone concentration because NO and NO2 lead to the loss and production of ozone, respectively. Figure 8 shows the annual average NO/NO2 ratio changes. In the aerosol-polluted areas, i.e. north China, the YRD, the PRD, the SCB and central China, the NO/NO2 ratio increased, with a highest value of 0.17. West China, south China (excluding the PRD) and north-east China were less influenced by ADE-POL, and the NO/NO2 ratio showed a small change. The observations in Germany (Melkonyan and Kuttler, 2012), Brazil (De Souza et al., 2017) and China (Han et al., 2011) have demonstrated that an increasing NO/NO2 ratio could consume more ozone and reduce ozone concentration. In ADE-MOI, JNO2 and JO1D were both increased due to the cloud amount and optical depth changes. Tang et al. (2003) found that the JNO2 was more sensitive to cloud than JO1D and most other photolysis rates and the decrease in cloud cover could lead to higher net ozone production below the cloud layer. Therefore, changes in the atmospheric water content and subsequent cloud changes could lead to local increases in surface ozone concentration.

These two opposite effects compete against each other, resulting in different ozone changes in different regions and seasons. Figure 9 presents the seasonal changes in the NO/NO2 ratio (representing the ADE-POL effect), photolysis rates (representing the ADE-MOI effect) and ozone concentration in the four selected regions and in the whole country. The increase in the NO/NO2 ratio dominates the ozone changes and diminished the surface ozone concentration in all seasons and regions, except for February in the YRD and SCB regions, when the ADE-MOI effect overwhelmed the ADE-POL effect. The magnitude of ozone percentage change appears to depend on the relative magnitude of the NO/NO2 ratio changes and the photolysis rates change. In northern cities, such as JJJ, the monthly variation in ozone changes showed a double-peak pattern, with the largest decline in spring and autumn, while in south China, the change in ozone only reached its largest reduction in winter. The latitudes of the YRD and the SCB are in between the latitudes of JJJ or the PRD, and therefore the seasonal patterns are not as clear as for JJJ or the PRD. In the YRD, the combined effect leads to ozone changes ranging from -5 to 0.07 ppb. Xing et al. (2017) found that the meteorology changes reduced the surface concentration of ozone in east China in January by 5–24 µgm-3 (2.33–11.19 ppb). The reason for the difference might be that they did not include the positive feedback of ADE-MOI when analysing meteorological effects. The reaction flux changes in Fig. S3 in the Supplement show that, on annual average, the combined effect of ADE-POL and ADE-MOI led to more ozone consumption than ozone production, suggesting that ADE-POL dominates. Figure 10 shows the spatial distribution of annual average ozone changes. The region with the highest ozone reduction is consistent with the region of the largest NO/NO2 ratio increase. Ozone concentration was found to decrease by 3.84 ppb (14.9 %), 2.45 ppb (8.7 %), 1.48 ppb (4.3 %) and 1.78 ppb (7.1 %) in JJJ, the YRD, the PRD and the SCB on annual average, and it decreased by around 2.01 ppb (6.2 %) averaged over the whole country.