CMA-CPSv3 is established based on a high-resolution version BCC-CSM2-HR fully coupled with four components (atmosphere, ocean, land, and sea ice) through fluxes of momentum, energy, and water at their interfaces (Wu et al., 2020b), and its coupled assimilation system (Liu et al., 2021). The atmospheric component in BCC-CSM2-HR is the Beijing Climate Center Atmospheric General Circulation Model version 3 (BCC-AGCM3, Wu et al., 2019), with a horizontal resolution of T266 triangular truncation (approximately 45 km) and 56 vertical hybrid sigma–pressure layers, with the top level at 0.1 hPa. The ocean component is the Modular Ocean Model Version 5 (MOM5, Griffies, 2012), with a horizontal resolution of 0.25° and 50 vertical levels. The land component is the Beijing Climate Center Atmosphere-Vegetation Interaction Model version 2 (BCC-AVIM2.0, Li et al., 2019), and the sea-ice component is the Sea Ice Simulator 5 (SIS 5, Delworth et al., 2006).

Recently, CMA-CPSv3 has been updated to CMA-CPSv4, in which the atmospheric component BCC-AGCM3 is updated to BCC-AGCM3-Chem, and the other three components (land, ocean, sea ice) in CMA-CPSv4 keep the same as those in CMA-CPSv3. BCC-AGCM3-Chem is based on the BCC-AGCM3 and coupled with interactive aerosols and their related atmospheric chemistry (here, other atmospheric chemistry predictions are not activated in order to improve computational efficiency) based on BCC-ESM1 (Wu et al., 2020b). Model physics in BCC-AGCM3-Chem has been further modified, including cloud parameterization, deep and shallow cumulus convection, boundary layer turbulence, and gravity wave schemes (Lu et al., 2020, 2021; Wu et al., 2021; Zheng et al., 2024). Those modifications are useful for improving the simulation of the Intertropical Convergence Zone (ITCZ), precipitation, tropical cyclones (TCs), the MJO, and the stratospheric QBO.

There are 13 major prognostic aerosol species involved in BCC-AGCM3-Chem, which are the same as those in BCC-ESM1 (Wu et al., 2020b), including four bins each for soil dust (DST01, DST02, DST03, DST04) and sea salt (SSLT01, SSLT02, SSLT03, SSLT04); two types of organic carbon (hydrophobic OC1 and hydrophilic OC2); two types of black carbon (hydrophobic BC1 and hydrophilic BC2) and sulfate (SO42-). These species are interactively treated through processes including emission, transport, gas-phase chemical reactions, secondary aerosol formation, gravitational settling, dry deposition, wet scavenging by clouds and precipitation, and aerosol effects on radiation, clouds, and precipitation. In the current version, the radiative transfer parameterization of BCC-AGCM3-Chem only accounts for the direct radiative effect of these 13 prognostic aerosol species, while other aerosol species, such as nitrate and ammonium, are not included as prognostic variables.

Aerosol-cloud interactions within the CMA-CPSv4 are manifested through the modulation of liquid cloud droplet number concentration (N_cdnc, cm⁻³) and effective droplet radius. The cloud droplet number concentration is parameterised using an empirical function proposed by Boucher and Lohmann (1995) and Quaas et al. (2006): 1Ncdnc=exp(5.1+0.41×ln(mss+moc+mso4+mNH4NO3)) Four types of hydrophilic aerosols influence cloud droplet number concentration, including bin 1 of sea salt, hydrophilic organic carbon, sulfates, and ammonium nitrate (NH₄NO₃). Among these, NH₄NO₃ is prescribed using a monthly mean climatology derived from NCAR CAM-Chem (Lamarque et al., 2012). Further details may be referred to the description in Wu et al. (2020b). In BCC-AGCM3 of previous CMA-CPSv3, those major aerosol mass concentrations for radiative transfer parameterization and N_cdnc estimates are prescribed from CMIP5-recommended historical data (Taylor et al., 2012).

In CMA-CPSv4, dust emission is interactively determined through coupling between the land component BCC-AVIM2.0 and BCC-AGCM3-Chem. BCC-AVIM2.0 provides land-surface conditions, and the emission flux is computed using a saltation–sand-blasting scheme that depends on wind friction velocity, soil moisture, and vegetation or snow cover (Zender et al., 2003). Sea salt emission is calculated online and parameterized as a function of the 10 m wind speed over the sea surface (Mahowald et al., 2006). Sulfates are formed from their precursors sulfur dioxide (SO₂) through gas-phase oxidation and aqueous-phase oxidation within cloud droplets. The primary sources of SO₂ include fuel combustion, industrial activities, and volcanic eruptions. The oxidation of dimethyl sulfide (DMS) produced by oceanic phytoplankton activity is also a major source of SO₂ (Table 1). Carbonaceous aerosols, including OC and BC, are categorised as hydrophobic and hydrophilic aerosols. Hydrophobic aerosols convert at a constant rate into hydrophilic aerosols. OC and BC originate primarily from fossil fuel combustion and biomass burning. The gas-phase chemistry in CMA-CPSv4 involves the following 6 reactions, with their reaction equations and corresponding rate coefficients listed in Table 1. For the oxidants OH and NO₃, the data are prescribed based on MOZART-4 climatological data, which are derived from different treatments within the BCC-ESM1 (Wu et al., 2020b).

In order to validate the performance of CMA-CPSv4 in simulating aerosols, we conducted a 20-year experiment from January 2001 to December 2020. The experiment was forced by atmospheric fields including horizontal winds, temperature, and specific humidity nudged toward 6-hourly ERA5 reanalysis fields (Hersbach et al. 2020, 2023a, b), and the ocean temperature in the ocean component of CMA-CPSv4 was nudged toward an ocean reanalysis produced by a newly developed coupled assimilation system based on CMA-CPSv3 (Liu et al., 2021).

The simulations were run with prescribed forcings based on CMIP6-recommended data (https://esgf-node.llnl.gov/search/input4mips/, last access: 4 January 2025) including: (1) greenhouse gas concentrations such as CO₂, N₂O, CH₄, CFC-11, and CFC-12 with monthly zonal-mean values, and ozone concentrations with monthly gridded values; (2) annual mean total solar irradiance derived from the CMIP6 solar forcing; and (3) stratospheric aerosols from volcanic eruptions. Aerosol emissions were also taken from CMIP6-recommended datasets, including biomass burning emissions (Van Marle et al., 2017) and anthropogenic and open burning emissions (Feng et al., 2020; Gidden et al., 2019; Hoesly et al., 2018).

The daily Modern-Era Retrospective Analysis for Research and Applications, version 2 aerosol reanalysis (MERRA-2; Buchard et al., 2017; Gelaro et al., 2017; Randles et al., 2017) and the multi-model mean of 10 Earth system models (ESMs) participating in CMIP6 (Eyring et al., 2016) are used to evaluate the CMA-CPSv4 simulation of five major aerosol species (dust, sea salt, sulfate, OC, and BC). MERRA-2 reanalysis is widely used for aerosol model evaluation (e.g., Buchard et al., 2017; Gui et al., 2026; Randles et al., 2017; Su et al., 2022; Turnock et al., 2020; Zhou et al., 2023). The CMIP6 ESMs include BCC-ESM1, CESM2-WACCM, EC-Earth3-AerChem, GFDL-ESM4, IPSL-CM5A2-INCA, MIROC-ES2L, MPI-ESM1.2-HAM, MRI-ESM2-0, NorESM2-LM, and UKESM1-0-LL. These model outputs are derived from the historical simulations of the first ensemble member (r1i1p1f1) for the period 1850–2014, obtained from the Earth System Grid Federation (ESGF, https://esgf-node.ornl.gov/search?project=input4MIPs, last access: 2 March 2025). The simulated data from 2001 to 2014 are selected and interpolated to a 1° × 1° grid for evaluation. As surface PM_2.5 is not directly available from all CMIP6 models, PM_2.5 concentrations are calculated following previous studies (Su et al., 2022; Turnock et al., 2020): 2PM2.5=BC+OC+SO4+0.1×DU+0.25×SS This approach assumes that 10 % of dust and 25 % of sea salt contribute to the PM_2.5 fraction and has been widely applied in CMIP5 and ACCMIP analyses (Silva et al., 2017; Turnock et al., 2020). In addition, two ground-based observational datasets are used to evaluate the simulated surface PM_2.5 concentrations. These include monthly mean PM_2.5 measurements from 28 stations of the East Asian Acid Deposition Monitoring Network (EANET, http://www.eanet.asia, last access: 4 March 2025, Ohizumi, 2023) and annual mean PM_2.5 concentrations from 364 urban monitoring sites operated by the China National Environmental Monitoring Centre (CNEMC, http://www.cnemc.cn, last access: 6 March 2025), and daily PM_2.5 observations from the CNEMC for the period 2014–2019. AOD observations are taken from the AErosol RObotic NETwork (AERONET, https://aeronet.gsfc.nasa.gov/new_web/, last access: 18 March 2025, Holben et al., 1998) using the cloud-screened and quality-assured Level 2.0 version 3.0 monthly product. AOD at 550 nm is derived from a log-linear interpolation between measurements at 440 and 675 nm (Dai et al., 2014; Schuster et al., 2006). In addition, monthly AOD products from the Moderate Resolution Imaging Spectroradiometer (MODIS, https://ladsweb.modaps.eosdis.nasa.gov/, last access: 20 March 2025, Platnick et al., 2017) and the Multi-angle Imaging SpectroRadiometer (MISR, https://l0dup05.larc.nasa.gov/L3Web/, last access: 20 March 2025, Diner et al., 1998) are also used for model evaluation.

The study also assesses the model's ability to reproduce monthly variations in PM_2.5 and AOD in different regions around the world according to definitions of the Hemispheric Transport of Air Pollution phase 3 (HTAP v3; Crippa et al., 2023). HTAP v3 was developed under the Task Force on Hemispheric Transport of Air Pollution (TF HTAP) as a comprehensive tool for simulating and analyzing the transboundary transport of air pollutants. In HTAP v3, the global land surface is divided into nine source regions: East Asia (EAS), the European Monitoring and Evaluation Programme domains (EMEP East Domain, EMEP_E; and EMEP West Domain, EMEP_W), Mexico/Central America/Caribbean (MCA), North America (NAM), South Asia (SAS), Southeast Asia (SEA), Southern and Eastern Mediterranean (SMD), and the Rest of the World (ROW; including South America, Southern Africa, and Australia). The regional definitions used in HTAP3-OPNS are publicly available (https://zenodo.org/records/12654038, last access: 20 March 2025, Butler, 2024).

Figure 1 shows the global spatial distributions of annual mean column-integrated concentrations of dust, sea salt, sulfate, OC and BC aerosols, averaged for the period 2001–2020 from CMA-CPSv4 simulations. The results are compared with the MERRA-2 reanalysis for the same period and with the multi-model ensemble mean of historical simulations for 2001–2014. High dust concentrations are mainly located over arid and semi-arid regions, including the Sahara, Arabian, and Taklamakan deserts, and their downwind areas (Fig. 1a–c). These areas are collectively referred to as the “dust belt” (Ginoux et al., 2012). CMA-CPSv4 reproduces the spatial pattern of dust concentrations in this region reasonably well but exhibits regional discrepancies in dust intensity. The model underestimates dust concentrations over the Taklamakan–Mongolia–North China region while overestimating them over North Africa and the Caspian Sea. Sea salt aerosols are primarily distributed over the Southern Ocean, the tropical South Indian Ocean, and the tropical Pacific Ocean, corresponding to regions of strong near-surface winds (Fig. 1d–f). MERRA-2 exhibits systematically higher concentrations than CMIP6 MME, which may reflect known biases, including an erroneous lake-masking algorithm, weak wet deposition and excessive coastal intrusion, all of which may contribute to an overall overestimation of sea salt concentrations (Buchard et al., 2017; Provençal et al., 2017; Randles et al., 2017). In comparison, CMA-CPSv4 simulates slightly lower concentrations in these regions, with values closer to those of the CMIP6 MME. The high concentrations of anthropogenic aerosols occur in high-emission areas associated with human activities. Sulfate aerosols, produced primarily through the oxidation of SO₂, are mainly concentrated over East Asia, South Asia, Europe, and the Middle East. The CMA-CPSv4 captures the distribution features of sulfate aerosols over land but overestimates their concentrations over the oceans (Fig. 1g–i). Carbonaceous aerosols (BC and OC), mainly from biomass burning and fossil fuel combustion, are concentrated in East Asia, Southeast Asia, South Asia, Africa, and South America. The simulated BC and OC concentrations are close to those in MERRA-2 and the multi-modal ensemble mean over land (Fig. 1j–o). The global mean values of dust (37.87 mg m⁻²) and sulfate (3.99 mg m⁻²) simulated by CMA-CPSv4 exceed those from MERRA-2 (33.92 and 3.03 mg m⁻², respectively) and the multi-model ensemble mean (26.86 and 3.18 mg m⁻², respectively). Such differences likely arise from variations in dry deposition parameterizations and emission inventories (Kong et al., 2025; Pleim et al., 2022). For sea salt, OC, and BC aerosols, the simulated global means lie between the corresponding MERRA-2 and the ensemble mean values. Spatial correlations between CMA-CPSv4 and MERRA-2 are 0.93 for dust, 0.92 for sea salt, 0.84 for sulfate, 0.84 for OC, and 0.90 for BC, reflecting consistency in their large-scale spatial distributions. Overall, CMA-CPSv4 reproduces the large-scale spatial distributions of the five aerosol species, although noticeable deviations remain in the simulated dust and sulfate concentrations.

Figure 2 presents the zonal mean column mass concentrations of dust, sea salt, sulfate, OC and BC aerosols simulated by CMA-CPSv4 (red line), in comparison with the MERRA-2 reanalysis (blue line) and the CMIP6 MME (black line). Compared to MERRA-2 and MME, CMA-CPSv4 reproduces the latitudinal patterns and peak locations of dust concentrations, although it overestimates values between 15° S and 30° N (Fig. 2a). At other latitudes, simulated dust concentrations are consistent with the multi-model ensemble mean but remain lower than MERRA-2 estimates. For sea salt aerosols, CMA-CPSv4 captures the two tropical peak concentrations present in MERRA-2, while underestimating sea salt concentrations near 60° S (Fig. 2b). The model reproduces the observed sulfate maximum near 25° N but exhibits positive biases in the tropics and Southern Hemisphere (SH) (Fig. 2c). The overestimation is likely linked to high model-simulated DMS emissions from oceanic phytoplankton activity, which enhance secondary sulfate formation or inadequate removal of sulfates over the ocean in the model. For carbonaceous aerosols, the simulated OC and BC concentrations decrease from the tropics toward the poles, consistent with MERRA-2 and the ensemble mean, but are overestimated in the 0°–30° N latitude band (Fig. 2d–e).

The vertical distributions of major aerosol species are shown in Fig. 3. Dust is concentrated primarily in the lower troposphere of the Northern Hemisphere (NH), particularly between 0° and 50° N, with concentrations exceeding 10 µg m⁻³ near 500 hPa (Fig. 3c). Sea salt aerosols are mainly confined to the lower troposphere below 700 hPa, particularly in the Southern Hemisphere (SH) (Fig. 3f). Anthropogenic aerosols, including sulfate and BC, are mainly concentrated below 500 hPa and within the latitudinal band of 20°–40° N, while OC shows enhanced concentrations near the equator and around 30° N (Fig. 3i, l, and o). Compared with MERRA-2 and the CMIP6 MME, CMA-CPSv4 reproduces the main lower-tropospheric distribution patterns of aerosol concentrations. However, over the tropical and subtropical region (0°–30° N), CMA-CPSv4 simulates aerosol layers extending to higher altitudes, particularly for dust, sea salt, and BC. Such differences are closely related to the stronger upward motion simulated over the tropical region (figure omitted), which enhances the vertical transport of boundary-layer aerosols into the upper troposphere. As a result, aerosols, including BC, are transported to higher altitudes over the 0°–30° N region in CMA-CPSv4 than in MERRA-2. In addition, the deep convection parameterization used in CMA-CPSv4 (Wu, 2012; Wu et al., 2019) may also contribute to transporting aerosols to higher altitudes through deep convective uplift and associated convective detrainment in the upper troposphere. Furthermore, because nudging constrains large-scale prognostic variables toward ERA5 reanalysis fields, it may suppress small-scale circulation variability and related precipitation processes, thereby weakening aerosol wet scavenging and allowing aerosols to persist at higher levels. These processes jointly contribute to the upward extension of aerosol layers in CMA-CPSv4 over the 0°–30° N region relative to MERRA-2.

Figure 4 presents monthly variations in regional surface PM_2.5 concentrations simulated by CMA-CPSv4, compared with the CMIP6 MME and MERRA-2 reanalysis. Over oceanic regions, MERRA-2 exhibits substantially higher surface PM_2.5 concentrations than both CMA-CPSv4 and the multi-model ensemble, primarily due to an overestimation of surface sea salt aerosols in MERRA-2 (Buchard et al., 2017; Provençal et al., 2017; Randles et al., 2017). In contrast, CMA-CPSv4 simulates PM_2.5 concentrations more consistent with the multi-model ensemble mean across the oceans (Fig. 4a). For East Asia, CMA-CPSv4 reproduces the spring PM_2.5 concentration peaks but underestimates the magnitude relative to both MERRA-2 and the ensemble mean (Fig. 4b). In the EMEP-East region, the model reasonably captures the seasonal variability but slightly overestimates concentrations during spring and autumn compared to the multi-model ensemble mean (Fig. 4c). Over North America, CMA-CPSv4 reproduces the annual cycle of surface PM_2.5 in the CMIP6 MME, including the summer (JJA) maximum, (Fig. 4f). CMA-CPSv4 also reproduces the monthly variability of PM_2.5 well in other regions – South Asia, Southeast Asia, the southern and eastern Mediterranean, and most land areas of the Southern Hemisphere – with correlation coefficients of 0.74, 0.88, 0.83, and 0.93, respectively (Fig. 4f, h, i and j). However, large biases remain over the EMEP-West region, Mexico, Central America, and the Caribbean (Fig. 4d and e). Overall, the annual PM_2.5 cycle in the CMIP6 MME is broadly consistent with MERRA-2 across most global regions. Nonetheless, both CMA-CPSv4 and the ensemble mean tend to underestimate PM_2.5 concentrations in the Northern Hemisphere (excluding East Asia) and over the oceans relative to MERRA-2. The annual cycle simulated by CMA-CPSv4 aligns more closely with the CMIP6 MME than with MERRA-2.

The comparisons of interannual variations in surface dust, anthropogenic aerosols, and PM_2.5 concentrations across four major aerosol source regions: East Asia, South Asia, the southern Mediterranean, and South Africa are shown in Fig. 5. For dust aerosols, the CMA-CPSv4 captures interannual variations across three regions: South Asia, the southern Mediterranean, and South Africa. The simulated temporal evolution and trends are consistent with MERRA-2, but systematic underestimations exist, particularly in East Asia and South Asia (Fig. 5a and b). The temporal variations of anthropogenic aerosols and PM_2.5 exhibit similar features. In East Asia, both anthropogenic aerosols and PM_2.5 show an increasing trend between 2001 and 2012, followed by a decline (Fig. 5e and i). In South Asia, both aerosols show statistically significant upward trends, with growth rates exceeding those in other regions (Fig. 5f and j). In contrast, trends in the southern Mediterranean and southern Africa are weaker (Fig. 5g, h, k and l). Simulations of both anthropogenic aerosols and PM_2.5 are correlated with MERRA-2, with correlation coefficients exceeding 0.71 in East Asia, South Asia, and South Africa. Overall, CMA-CPSv4 reasonably reproduces the interannual variations and trends of surface dust, anthropogenic aerosols, and PM_2.5 across these major aerosol source regions, though simulated concentrations still exhibit deviations.

Figure 6 presents the spatial distribution of annual mean surface PM_2.5 concentrations at 392 monitoring sites across East Asia, derived from CNEMC and EANET observations, MERRA-2 reanalysis, the CMIP6 MME, and CMA-CPSv4 simulations averaged over 2001–2020. Compared with observations from CNEMC and EANET, MERRA-2, the CMIP6 MME, and CMA-CPSv4 all underestimate annual mean PM_2.5 concentrations across East Asia (Fig. 6b, c and d). The CMA-CPSv4 results are broadly consistent with the CMIP6 MME but underestimate the observed PM_2.5 values by up to 20 µg m⁻³ in high-concentration regions of eastern China (Fig. 6c and d). In addition, the model also distinctly underestimates PM_2.5 concentrations in the Taklamakan–Mongolia–North China region, primarily due to its underestimation of dust concentrations in this area (Fig. 6a and d).

Figure 7 further compares simulated and observed PM_2.5 concentrations through scatter plots of CMA-CPSv4 against ground-based observations, MERRA-2, and the CMIP6 MME. Relative to the CMIP6 MME and MERRA-2, CMA-CPSv4 reproduces the spatial variability of PM_2.5 well, with correlation coefficients of 0.95 and 0.90, respectively (Fig. 7b, c). The simulated values are generally distributed close to the 1:1 line, although CMA-CPSv4 tends to be slightly lower than MERRA-2. In contrast, CMA-CPSv4 substantially underestimates PM_2.5 concentrations compared to site observations, with simulated values averaging only about half of the observed levels, indicating a low bias in model performance over East Asia. Such underestimation of PM_2.5 concentrations is a common issue among CMIP6 models (Ren et al., 2024; Turnock et al., 2020; Zhang et al., 2012). The differences between CMIP6 models and observations may partly result from uncertainties in the partitioning of natural aerosols into the PM_2.5 fraction, as well as the omission of secondary organic aerosols, ammonium, and nitrate in PM_2.5 calculations (Shrivastava et al., 2017; Zheng et al., 2015).

This section further assesses the model performance in reproducing the global distribution and seasonal variations of AOD. The spatial pattern of AOD is modulated by aerosol composition, concentration, and the influence of clouds. Figure 8 presents the spatial distribution of the global annual mean AOD derived from AERONET, MERRA-2, MISR, MODIS, CMIP6 MME and CMA-CPSv4 simulations, averaged over the period 2001–2020. On a global scale, high AOD areas are mainly located in North Africa, the Middle East, South Asia, and East Asia within the Northern Hemisphere (Fig. 8a–f). In these source regions, the simulated AOD magnitudes in the high-value areas are comparable to those from AERONET, MODIS and CMIP6 MME (Fig. 8a, c, e and f), while slightly exceeding the values from MERRA-2 and MISR (Fig. 8b and d). The simulated spatial pattern of AOD over North America is broadly consistent with AERONET observations and CMIP6 MME. Over oceanic regions, the CMA-CPSv4 simulation shows similar spatial distributions to MERRA-2 data and CMIP6 MME. High AOD values observed by MISR, MERRA-2, and CMIP6 MME near 60° S, primarily associated with high sea salt concentrations, are also reproduced by CMA-CPSv4 (Fig. 8b, d, e and f). Despite these agreements, certain regional biases remain. For instance, the model overestimates AOD over Europe, western North Africa, and southern China relative to the observations. Overall, the simulated AOD from CMA-CPSv4 captures the principal spatial distribution features consistent with AERONET, satellite datasets, MERRA-2 and CMIP6 MME on the global scale, with a correlation coefficient of 0.81 with AERONET observations. The global mean AOD simulated by CMA-CPSv4 (0.1271) is slightly higher than that from MERRA-2 (0.1191), CMIP6 MME (0.117) and MISR (0.1186), which is within the uncertainty range (±0.05) of the MISR AOD detection (Kahn et al., 1998).

The comparisons of monthly mean AOD simulated by CMA-CPSv4 with observations from AERONET, MERRA-2 reanalysis, and CMIP6 MME are presented in Fig. 9. A total of 5275 data pairs is analyzed, including 2988 data pairs for the comparison with the CMIP6 MME. The simulated AOD intensity closely matches the AERONET AOD intensity at the vast majority of locations, although slight underestimation occurs at a few sites (Fig. 9a). The underestimation likely results from spatial averaging of high-emission sources within coarse-resolution grid cells, which smooths localized emission peaks in the simulation results. Compared with MERRA-2 and the CMIP6 MME, CMA-CPSv4 exhibits higher correlations and lower RMSEs (Fig. 9b, c). The correlation coefficient between simulated and AERONET AOD is 0.78, increasing to 0.84 and 0.88 for MERRA-2 and the CMIP6 MME, respectively. The corresponding root mean square errors (RMSEs) decrease to 0.071 and 0.065 for MERRA-2 and CMIP6 MME, respectively, compared with 0.095 for AERONET. These results indicate that the simulated AOD is more consistent with MERRA-2 and the CMIP6 MME than with AERONET observations. Overall, the simulated AOD can capture the observed variability, with correlation coefficients exceeding 0.78 for all three reference datasets.

Figure 10 presents the monthly variations of simulated AOD from CMA-CPSv4 compared with AERONET observations, MERRA-2 reanalysis and CMIP6 MME across different regions. In general, the simulated AOD is within the error range of the observations. The model reproduces the annual cycle of AOD reasonably well in several regions, particularly over North America, the southern and eastern Mediterranean, the ocean, and Southern Hemisphere land areas, with correlation coefficients (R) of 0.95, 0.94, 0.86, and 0.83, respectively (Fig. 10a, f, g and j). The CMIP6 MME is also relatively close to AERONET and MERRA-2 in these regions. However, CMA-CPSv4 notably overestimates AOD during June–August (JJA) in most regions, including Mexico, Central America, the Caribbean, South Asia, Southeast Asia, and the EMEP-West region (Fig. 10d, e, h and i). Moreover, the simulated AOD peaks in these regions occur later in the CMA-CPSv4 (summer) compared to observations (spring), likely due to an overestimation of summertime sulfate aerosol concentrations by CMA-CPSv4. In East Asia, CMA-CPSv4 captures the general seasonal variation patterns during spring and summer but underestimates their intensity, while AOD values are overestimated during autumn (Fig. 10b). In Southeast Asia and the EMEP-East region, the model fails to reproduce the observed amplitude of seasonal variation, showing weaker seasonality compared with AERONET and MERRA-2 (Fig.  10d and i). Large differences between the CMIP6 MME and both AERONET and MERRA-2 are also evident over East Asia, the EMEP-East region, South Asia, and Southeast Asia (Fig. 10b, c, h and i), suggesting that accurately simulating aerosol seasonal variations in these regions remains challenging for current ESMs.

In the previous sections, the performance of CMA-CPSv4 in reproducing aerosol concentrations and AOD distributions was systematically evaluated against observations, the MERRA-2 reanalysis and the CMIP6 MME. To further assess model performance during high-pollution conditions, this section presents an evaluation of PM_2.5 pollution events in terms of their spatial occurrence frequency distributions and temporal evolution at representative sites. In addition, we selected two severe dust pollution events to evaluate the model's ability to reproduce such pollution events.

Figure 11 shows the spatial distributions of PM_2.5 pollution-event occurrence frequencies over China during 2014–2019 from CNEMC observations, MERRA-2 reanalysis, and CMA-CPSv4 simulations. For mild pollution events (35 µg m⁻³ ≤ PM_2.5 mass < 75 µg m⁻³), such events occur frequently across most regions of China, with occurrence frequencies generally exceeding 10 % and reaching more than 30 % over the North China Plain, the Yangtze River Delta, and the Sichuan Basin (Fig. 11a). Both MERRA-2 and CMA-CPSv4 reproduce the major spatial distribution features, particularly over the Yangtze River Delta, the Sichuan Basin, and central China, although occurrence frequencies are slightly higher in these regions (Fig. 11b, c). Similar spatial patterns are found in MERRA-2 and CMA-CPSv4 for mild pollution events. Negative deviations relative to observations are mainly concentrated over the Taklamakan Desert, Gansu, Inner Mongolia, and parts of southwestern China (Fig. 11c), likely associated with the underestimation of dust concentrations in CMA-CPSv4. Approximately 35.6 % and 48.0 % of stations in CMA-CPSv4 and MERRA-2, respectively, exhibit absolute deviations within the 10 % error threshold. For moderate-to-severe pollution events (PM_2.5 mass ≥ 75 µg m⁻³), underestimations are mainly found over the North China Plain, northeastern China, and southwestern China (Fig. 11e and f). Relative to MERRA-2, CMA-CPSv4 shows smaller deviations at some stations over the North China Plain, although underestimations persist over the Taklamakan Desert, Gansu, and Inner Mongolia (Fig. 11f). Meanwhile, 60.9 % and 63.8 % of stations in MERRA-2 and CMA-CPSv4 satisfy the 10 % absolute-deviation criterion relative to observations. These results indicate that both datasets tend to underestimate the occurrence frequency of pollution events.

To further evaluate the model's performance in reproducing the temporal evolution of pollution episodes, the Beijing monitoring station on the North China Plain is selected as a representative site. This region experiences frequent pollution episodes in winter, primarily associated with enhanced anthropogenic aerosol emissions during this season. Figure 12 shows the daily evolution of PM_2.5 concentrations at the Beijing site between December 2014 and December 2019 from CNEMC observations, MERRA-2 reanalysis, and CMA-CPSv4 simulation. The temporal variations simulated by CMA-CPSv4 are broadly consistent with those from MERRA-2, including the timing, peak intensity, and duration of pollution episodes. Although both datasets significantly underestimated PM_2.5 concentrations during pollution episodes compared to observations, the timing and persistence of most pollution peaks are generally captured (Fig. 12a–f). Among the 31 pollution episodes identified during the study period, the timing of peak concentrations is reproduced in 27 cases, while the remaining four cases show peak occurrences advanced by one day. These results indicate that CMA-CPSv4 captures the temporal evolution and peak timing of pollution episodes in the North China Plain. The lower simulated PM_2.5 concentrations during pollution episodes are consistent with the general underestimation discussed in Sect. 3.5.

Here, we further examine the model's ability to reproduce severe dust pollution events. Figure 13 presents Hovmöller diagrams of dust mass concentration along 39° N from 1 to 30 May 2011. According to MERRA-2, two major dust transport events occurred during 1–6 May and 14–21 May 2011, respectively, with a minor dust episode observed around 9–10 May (Fig. 13a). The first major event originated near 100° E and propagated eastward to approximately 130° E, while the weaker mid-month event extended to about 125° E. The second major dust event (14–21 May) exhibited the highest intensity and widest transport extent, with dust plumes reaching the western Pacific around 140° E. CMA-CPSv4 successfully reproduces the temporal evolution and spatial progression of these dust plumes – from initial emissions over western China to their subsequent eastward transport – although the simulated dust intensity is slightly underestimated (Fig. 13b).

Figure 14 shows the spatial distribution of dust mass concentrations simulated by CMA-CPSv4 for the dust event over China during 15–21 May 2011, in comparison with MERRA-2 reanalysis data. Overall, the spatial pattern of dust concentrations reproduced by CMA-CPSv4 is consistent with that of MERRA-2. Initially, a dust plume develops over northwestern Inner Mongolia (Fig. 14a, b). Under the influence of large-scale circulation, the plume intensifies and propagates southeastward, affecting western and eastern Inner Mongolia, Northwest China, North China, Central China, and East China (Fig. 14c–f). Subsequently, the plume continues eastward and dissipates over the Pacific Ocean (Fig. 14g, h). The model successfully captures the temporal evolution and transport pathway of the dust plumes, indicating that CMA-CPSv4 can realistically reproduce the major features of dust emission and transport during this severe event.

Figure 15 compares the temporal evolution of dust concentrations at the Shandan Station during the 12–21 May 2011 dust event as simulated by CMA-CPSv4. The observed dust concentrations gradually increased from 12 May, peaked on 17 May, and were followed by a rapid decline reaching a minimum on 21 May. CMA-CPSv4 successfully reproduces the timing of the peak and the trends during this event, although it generally underestimates the magnitude relative to MERRA-2. Meteorological conditions on 16 May, the day preceding the peak, the largest differences in surface air and 500 hPa temperatures and maximum surface wind speed occurred in Shandan, resulting in unstable stratification and enhanced dust uplift and transport. CMA-CPSv4 captures the temperature gradient and wind speed variations reasonably well, though with a slight overestimation of surface winds. Overall, the model demonstrates good performance in reproducing the temporal features and meteorological conditions of this dust pollution event.