Xiaole SU, Tongwen WU*, Jie ZHANG, Yong ZHANG, Junli JIN, Qing ZHOU, Fang ZHANG,Yiming LIU, Yumeng ZHOU, Lin ZHANG, Steven T. TURNOCK, and Kalli FURTADO

1 Chinese Academy of Meteorological Sciences, China Meteorological Administration, Beijing 100081, China

2 Beijing Climate Center, China Meteorological Administration, Beijing 100081, China

3 Meteorological Observation Center, China Meteorological Administration, Beijing 100081, China

4 Laboratory for Climate and Ocean–Atmosphere Studies, Department of Atmospheric and Oceanic Sciences, School of Physics,Peking University, Beijing 100871, China

5 Met Office, Hadley Centre, Exeter EX1 3PB, United Kingdom

6 University of Leeds Met Office Strategic (LUMOS) Research Group, School of Earth and Environment,University of Leeds, Leeds LS2 9JT, United Kingdom

ABSTRACT This study assesses the ability of 10 Earth System Models (ESMs) that participated in the Coupled Model Intercomparison Project Phase 6 (CMIP6) to reproduce the present-day inhalable particles with diameters less than 2.5 micrometers (PM2.5) over Asia and discusses the uncertainty. PM2.5 accounts for more than 30% of the surface total aerosol (fine and coarse) concentration over Asia, except for central Asia. The simulated spatial distributions of PM2.5 and its components, averaged from 2005 to 2020, are consistent with the Modern-Era Retrospective Analysis for Research and Applications version 2 (MERRA-2) reanalysis. They are characterized by the high PM2.5 concentrations in eastern China and northern India where anthropogenic components such as sulfate and organic aerosol dominate, and in northwestern China where the mineral dust in PM2.5 fine particles (PM2.5DU) dominates. The present-day multimodel mean (MME) PM2.5 concentrations slightly underestimate ground-based observations in the same period of 2014–2019, although observations are affected by the limited coverage of observation sites and the urban areas.Those model biases partly come from other aerosols (such as nitrate and ammonium) not involved in our analyses,and also are contributed by large uncertainty in PM2.5 simulations on local scale among ESMs. The model uncertainties over East Asia are mainly attributed to sulfate and PM2.5DU; over South Asia, they are attributed to sulfate, organic aerosol, and PM2.5DU; over Southeast Asia, they are attributed to sea salt in PM2.5 fine particles (PM2.5SS); and over central Asia, they are attributed to PM2.5DU. They are mainly caused by the different representations of aerosols within individual ESMs including the representation of aerosol size distributions, dynamic transport, and physical and chemistry mechanisms.

Key words: PM2.5, Asia, Coupled Model Intercomparison Project Phase 6 (CMIP6), Earth System Models (ESMs)

1.Introduction

Aerosol is a multiphase system composed of solid particles and liquid droplets, suspended in a gaseous carrier phase (e.g., air). Atmospheric aerosols can include minerals (e.g., silicates) originating from soils and rocks,carbonaceous components (black carbon and organic carbon), sulfates, nitrates, ammonium salts, sea salts, and biogenic components (Wang and Zhang, 2001; Zhang Y.et al., 2019). Through either direct (Coakley et al., 1983;Jacobson, 2001; Bond et al., 2013; Li et al., 2017) or indirect effects on atmospheric radiation (Charlson et al.,1992; Guo et al., 2018; Liu et al., 2021), aerosols are well recognized to significantly influence weather and climate at regional and global scales (Menon et al., 2002;Lau et al., 2006; Zhang et al., 2007; Tosca et al., 2010;Bollasina et al., 2011; Li et al., 2011; Wang et al., 2011,2013; Hwang et al., 2013; Wu G. X. et al., 2016; Zhang et al., 2021). Aerosols can also cause serious environmental problems such as fog, haze, photochemical smog,and acid rain, with significant impacts on the hydrological cycle, new energy development, agricultural production,and transportation (Ramanathan et al., 2001; Haywood et al., 2011; Singh et al., 2017; Sweerts et al., 2019). Fine particulate matters with particle diameters less than 2.5 μm, commonly termed PM2.5, are generally thought of as one of the main causes of air pollution and have an adverse effect on human health. According to the Global Burden of Disease (GBD) 2010 comparative risk assessment (Lim et al., 2012), roughly 3.2 million deaths per year are attributable to ambient PM2.5. Understanding and predicting PM2.5and its spatial and temporal variations are therefore vital for reducing mortality and other impacts on the environment (Apte et al., 2015).

With the development of Earth System Models(ESMs), the importance of coupling between multiple components of the earth system, including atmosphere,ocean, land, and sea ice, has gradually been recognized,and increasingly improved within these ESMs. ESMs have become an important tool to simulate and forecast global aerosols (Collins et al., 2017) and can not only fill the gaps between historical observations, but also estimate the trends of aerosols in the future, and thus provide a basis for assessing the evolution of air pollution in both the past and future. The performance of ESMs to reproduce the observed aerosols is an important issue for climate modeling communities. In fact, the Atmospheric Chemistry and Climate Model Intercomparison Project(ACCMIP) was endorsed by the Coupled Model Intercomparison Project Phase 5 (CMIP5), and tended to focus on the atmospheric chemistry (Lamarque et al.,2013), with only a few models providing the simulation results for aerosols (Collins et al., 2017). The Aerosols and Chemistry Model Intercomparison Project (Aer-ChemMIP; Collins et al., 2017), part of the Coupled Model Intercomparison Project Phase 6 (CMIP6; Eyring et al., 2016), provides an opportunity to understand the performance of the latest ESMs in simulating aerosols.There are few relevant assessments on the performance of CMIP6 ESMs in simulating aerosols (Mulcahy et al.,2020; Wu et al., 2020). They show that most of the current generation of ESMs such as BCC-ESM1 and UKESM1 can reproduce the global spatial distributions of most aerosol components (e.g., sulfate) concentrations, although there are some model biases for certain components.

It is important to understand the evolution of groundlevel PM2.5over Asia as it is one of the most heavily polluted regions on the globe, and has the highest mortality rate attributed to atmospheric pollution (Apte et al.,2015). Previous studies show that most of the CMIP6 ESMs can capture the spatial distributions of surface PM2.5concentrations across the globe but underestimate the absolute magnitude (Turnock et al., 2020). However,the ability of the CMIP6 ESMs to simulate PM2.5in Asia has not been carefully explored so far largely due to the lack of ground-based surface aerosol observations in Asia. In addition, the various components of PM2.5have seldom been utilized in previous studies, leading to the differences among models being poorly understood.

Here, simulations of surface PM2.5and its component concentrations from 10 CMIP6 ESMs are evaluated in detail against observations from surface sites over Asia.Based on the ratio of PM2.5to main aerosol mass and the relative contributions of each component to PM2.5, differences among models are revealed. The remaining parts of this manuscript are as follows. The research data and methods are presented in Section 2. In Section 3, we assess the ability of the CMIP6 ESMs to simulate the spatial distribution of PM2.5and its main components in Asia. In Section 4, we analyze their model-spread among 10 ESMs. Uncertainties in evaluating PM2.5concentrations are discussed in Section 5. A summary is given in Section 6.

2.Data and methods

The monthly mean PM2.5components, including sulfate, organic aerosol (OA), black carbon (BC), dust,and sea salt, from 10 ESMs participated in CMIP6 are employed in this study. The model information is described in Table 1, and all the model data can be freely downloaded from the Earth System Grid Federation(ESGF) nodes (https://esgf-node.llnl.gov/search/cmip6/).All the models use the same anthropogenic emission inventory from the Community Emissions Data System(CEDS; Hoesly et al., 2018; http://www.globalchange.umd.edu/ceds/ceds-cmip6-data/) and their own schemes for simulating natural emissions such as dust and sea salt aerosols, which have different representations of the aerosol size distribution (Collins et al., 2017). The model data are obtained from the CMIP6 historical experiments(Eyring et al., 2016) before 2015 and from the SSP370 experiments in AerChemMIP (Collins et al., 2017) afterward.

Not all CMIP6 ESMs provide PM2.5concentrations,even for some ESMs with available PM2.5, they use different methods to calculate it. In order to uniformly evaluate the ability of ESMs to simulate PM2.5, it is necessary to find a consistent method to calculate PM2.5.Therefore, following the methods used in other studies(Silva et al., 2013; Turnock et al., 2020), the formula used to estimate PM2.5mass concentrations from the ESMs data is expressed as

Table 1. CMIP6 earth system models used in this study

where BC, OA, SO4, DU, and SS represent the black carbon (CMIP6 diagnostic identifier: mmrbc), organic aerosol (mmroa), sulfate (mmrso4), dust (mmrdust), and sea salt (mmrss) mass mixing ratio (kg kg?1), respectively.All the aerosol mass concentrations in the lowest layer of each ESM are taken as the near surface values from simulations in this work. The particles for BC, OA, and SO4aerosols are generally less than 2.5 μm in diameter.

In Eq. (1), 10% and 25% of dust and sea salt particles are assumed to be present within the fine size fraction of less than 2.5 μm in diameter. We validated this assumption for dust and sea salt from additional BCC-ESM1 simulations, which provided output across four-size bins of dust (DST01: 0.1–1.0 μm, DST02: 1.0–2.5 μm,DST03: 2.5–5.0 μm, and DST04: 5.0–10 μm) and sea salt (SSLT01: 0.2–1.0 μm, SSLT02: 1.0–3.0 μm,SSLT03: 3.0–10 μm, and SSLT04: 10–20 μm) aerosols(Wu et al., 2020). Only the ESGF provides total aerosol mass mixing ratios so we only have access to full size resolved aerosol data from BCC-ESM1. As shown in Fig.1, the estimated PM2.5fine particles concentrations for dust (hereafter PM2.5DU) and sea salt (PM2.5SS) from the Eq. (1) are nearly consistent to that from the original BCC-ESM1 simulations (fine size fraction less than 2.5 μm in diameter calculated by summing by DST01,DST02, SSLT01, and SSLT02, respectively).

Fig. 1. Annual mean of near surface PM2.5DU and PM2.5SS concentrations (μg m?3) in Asia (5°–55°N, 70°–140°E) during 2005–2020 from BCC-ESM1 simulations. (a, c) PM2.5DU and (b, d) PM2.5SS, and (a, b) estimation and (c, d) original.

To evaluate the present-day PM2.5climatology in ESMs, the following ground-based observations are used: monthly mean surface PM2.5observations during 2014–2019 at 25 sites in Asia from the Acid Deposition Monitoring Network in East Asia (hereafter EANET data; http://www.eanet.asia, accessed on 16 December 2020) and 348 urban sites in China available from the Chinese National Environmental Monitoring Center(hereafter CNEMC data; http://www.cnemc.cn, accessed on 16 December 2020). The CNEMC data have been used in previous studies (Wei et al., 2019; Wei et al.,2020). In order to examine the observation uncertainty due to the impact of urban effects, monthly mean PM2.5concentrations at two atmospheric background stations from the Meteorological Observation Center, China Meteorological Administration (hereafter CMA data; Zhang et al., 2020) are compared with the nearby urban sites from CNEMC data, as well as from a pair of urban and suburban ground-based observations in Thailand(Pathumwan and KlongHa) from the Asia–Pacific Aerosol Database (APAD; Cohen et al., 2015). The geographic distributions of all the observation sites and division of Asian subregions used in this study are shown in Fig. 2.

Fig. 2. Locations of observation sites in Asia (5°–55°N, 70°–140°E)from EANET (blue triangles, 25 sites), CNEMC (red circles, 348 urban sites), CMA (green circles, 2 background stations), and APAD (purple hollow squares, 2 adjacent sites). The dashed areas represent the various parts of Asia, including Central Asia (CA), East Asia (EA), South Asia (SA), and Southeast Asia (SEA).

Considering the sparsely covered and unevenly distributed ground-based observation, the Modern-Era Retrospective Analysis for Research and Applications, version 2 (MERRA-2) data with a high-resolution (0.5° ×0.625°) assimilation data product (including sulfate, organic aerosols, black carbon, dust, and sea salt) developed by combining satellite observations with the Goddard Earth Observing System atmospheric model and atmosphere data assimilation system (Buchard et al.,2016; Randles et al., 2017) are further used. The MERRA-2 data are widely used by many studies in evaluation of aerosols simulations (Turnock et al., 2020; Ukhov et al., 2020; Li et al., 2021; Zhao et al., 2021). For intercomparison between ESMs and MERRA-2, we derive the monthly MERRA-2 PM2.5data from 2005 to 2020 using Eq. (1), on the basis of the monthly sulfate,organic aerosols, black carbon, and total mass of dust and sea salt aerosols mass data that are directly downloaded from the website (https://gmao.gsfc.nasa.gov/reanalysis/MERRA-2/data_access/, accessed on 16 December 2020). In this study, all model data were interpolated to the same horizontal resolution of 0.5° × 0.625° latitude/longitude grids as in MERRA-2, and onto the site locations when compared with the ground-based observations.

3.Present-day climate of PM2.5 and its components in Asia

3.1 PM2.5 concentrations

In this section, we will focus on the spatial features of present-day climate mean PM2.5from 2005 to 2020. Figure 3 shows the percentage contribution of PM2.5to the total aerosol (fine and coarse) concentration in Asia, including sulfate, OA, BC, and all particle sizes of dust and sea salt. The results from MERRA-2 (Fig. 3l) show a relatively high proportion of PM2.5over East Asia and Southeast Asia and the contribution is up to 60%–80%over the southeastern coast of China. Central Asia is an arid or semi-arid region and has the lowest proportion(less than 30%) of PM2.5, where mineral dust is generally the main source of aerosols, and coarse particles dominate. For the multi-model mean (MME, Fig. 3k), the PM2.5ratio is in overall a good agreement with MERRA-2, except for MIROC-ES2L (Fig. 3f). MIROC-ES2L shows the largest proportion of fine particulate matter in eastern China, which is about 20% higher than in MME and MERRA-2.

Figure 4a shows the spatial distribution of present-day mean of surface PM2.5concentrations in the 373 groundbased observations from CNEMC and EANET, averaged for 2014–2019. Annual mean surface PM2.5concentrations in most parts of eastern China can be over 40 μg m?3, and the highest values are mainly centered over the Beijing–Tianjin–Hebei region where PM2.5concentration may be over 60 μg m?3. High annual mean PM2.5concentrations are also present over northwestern China,mainly contributed by mineral dust. In the area south of 25°N, the annual mean PM2.5concentrations are generally smaller, which may be caused by strong wet deposition and lower emissions. Japan and Korea are also regions with values of annual mean PM2.5concentrations less than 20 μg m?3. Figures 4b–k show the point-topoint comparisons between 10 ESMs simulations separately with 373 ground-based observations in the same period from 2014 to 2019. They illustrate that most models underestimate the observations, although all ESMs show high spatial correlations of 0.52–0.74 and 0.69 for MME (Fig. 4l). The underestimation of PM2.5concentrations by CMIP6 models in this study partly comes from the use of the approximate method to calculate PM2.5by Eq. (1), in which nitrateand ammoniumaerosols are not involved. Those underestimations also exist in CMIP5 models (Wu J. et al., 2016; Liu et al.,2017).

As shown in Fig. 4m, the MERRA-2 data also underestimate the observed PM2.5concentrations at 373 sites.Nevertheless, MERRA-2 can provide the overall spatial distribution of PM2.5in Asia with better temporal and spatial coverage and compensate for the gaps not covered by site observations. As shown in Fig. 5l, the spatial distribution of annual mean surface PM2.5concentrations averaged for 2005 to 2020 from MERRA-2 is similar to that from ground-based observations (Fig. 4a). Except for the two regions with high surface PM2.5concentrations in eastern China and northwestern China that can be found from ground-based observations, MERRA-2 (Fig. 5l)also shows a third region of high-concentration centered in northern India where there are high local emissions and the Himalayas plays a large role in preventing dispersal of aerosols (Shi et al., 2018). The PM2.5concentrations are less than 5 μg m?3over the Tibetan Plateau(about 26°–39°N, 73°–104°E) and Mongolia Plateau(about 37°–53°N, 87°–122°E), where human activities are weak.

Fig. 3. The 2005–2020 mean PM2.5 ratios (%) to main aerosol (including all particle sizes of dust and sea salt, sulfate, organic aerosol, and black carbon) in Asia (5°–55°N, 70°–140°E) for (a–j) the 10 ESMs, (k) their MME, and (l) MERRA-2.

The main spatial features of surface PM2.5concentrations are generally well captured by the ESMs (Fig.5a–k) in comparison with MERRA-2 (Fig. 5l), except over the offshore area where MERRA-2 data overestimated sea salt as pointed out in Buchard et al. (2017).However, there exists a large diversity among models,especially over the three PM2.5centers (eastern China,northern India, and northwestern China and Mongolia,Fig. 6a). The amplitude of model-spread (that is denoted by the standard deviation of simulated PM2.5concentration among 10 ESMs in the study) over northwestern China and Mongolia are close to the MME regional PM2.5concentration (Fig. 6b). Specifically, CESM2-WACCM (Fig. 5b) overestimates PM2.5in Taklimakan desert of central Xinjiang (> 60 μg m?3), and MRIESM2-0 (Fig. 5h) has an abnormally high-value center in the Mongolian Plateau. The dominant species of PM2.5vary with regions as well as the one responsible for the model-spread in PM2.5simulation, which will be discussed in detail in Section 4.

Fig. 4. (a) Averaged annual (2014–2019) mean surface PM2.5 concentrations (μg m?3) for 373 sites from EANET (triangles, 25 sites) and CNEMC (circles, 348 urban sites) in Asia. (b–m) Scatterplots of surface PM2.5 concentrations (μg m?3) for each ESMs and their MME, and MERRA-2, separately, comparing to the observations from EANET and CNEMC sites during the same period. RMSE stands for root-meansquare error, and COR for correlation coefficient. The grey lines represent the 1 : 1, 1 : 2, and 2 : 1 lines, respectively.

3.2 Main components of PM2.5 concentrations

Sulfate, OA, and BC are the main PM2.5aerosols from anthropogenic emissions in Asia and are the main PM2.5species over eastern China and northern India (Figs.7–9). In MERRA-2, sulfate (Fig. 7l) and BC (Fig. 9l)concentrations in eastern China are higher than those in northern India, whereas the spatial distribution for OA shows the opposite (Fig. 8l). The MME can generally reproduce the spatial distributions for sulfate (Fig. 7k), OA(Fig. 8k), and BC (Fig. 9k) although their magnitudes are underestimated for sulfate but overestimated for OA and BC. There are significant differences in the simulations of sulfate and OA among various ESMs. MRI-ESM2-0(Fig. 7h) has the highest concentration of sulfate in southeastern China, while IPSL-CM5A2-INCA (Fig. 7e)has the lowest sulfate concentrations. CESM2-WACCM(Fig. 8b) and UKESM1-0-LL (Fig. 8j) have larger concentrations of OA than other ESMs, which may be caused by different volatile organic compounds (VOC)and secondary organic aerosol (SOA) formation mechanisms in the ESMs. UKESM1-0-LL also shows the largest BC concentration than the others (Fig. 9j).

Fig. 5. Averaged annual (2005–2020) mean surface PM2.5 concentrations (μg m?3) in Asia (5°–55°N, 70°–140°E) from (a–j) 10 ESMs, (k) their MME, and (l) MERRA-2.

PM2.5DU and PM2.5SS are the natural components in PM2.5. As shown in Fig. 10, PM2.5DU is responsible for the PM2.5center (Fig. 5) over northwestern China and Mongolia. The PM2.5DU concentration from MME (Fig.10k) is similar to that from MERRA-2 (Fig. 10l). But there are large differences in PM2.5DU simulations among 10 ESMs. CESM2-WACCM (Fig. 10b) and GFDL-ESM4 (Fig. 10d) simulated larger PM2.5DU concentrations than other models. Moreover, the PM2.5DU in MIROC-ES2L (Fig. 10f) is much smaller than MERRA-2 (Fig. 10l), with PM2.5DU differences up to 20 μg m?3.In MRI-ESM2-0 (Fig. 10h), the high PM2.5DU center extends eastward to northern China and the amplitude of PM2.5DU is about twice of that in the east, which is not evident in MERRA-2 (Fig. 10l). In addition, MPI-ESM-1-2-HAM (Fig. 10g) simulated excessive amount of PM2.5DU in northern Tibetan Plateau, which is distinctive from other models. PM2.5SS is another important natural aerosol mainly distributed over oceans and coastal regions. The PM2.5SS concentration over land is lower than the other species in PM2.5, and the differences among ESMs are generally small (Fig. 11). Due to the known overestimation of sea salt in MERRA-2 (Buchard et al., 2017), there are significant differences between the MME and MERRA-2 (Figs. 11k, l).

Fig. 6. (a) Model-spread (μg m?3) among the 10 ESMs and (b) the ratio (%) of model-spread to MME for annual mean of surface PM2.5 concentration during 2005–2020.

4.Uncertainties in simulated PM2.5 concentrations from ESMs

4.1 Uncertainty in the anthropogenic and natural PM2.5 species

Figure 12 shows the model-spread among 10 ESMs for main anthropogenic components of PM2.5, sulfate,OA, and BC. The regions of large model-spread are evident over eastern China, northern India, and Sichuan Basin, and the main anthropogenic emission centers in Asia. All the ESMs used the same anthropogenic emissions inventory (Hoesly et al., 2018). Large modelspread for anthropogenic aerosols in individual ESMs thus mainly comes from the different way that individual models represent chemical and physical processes relevant for aerosols including dynamic transport, dry deposition, gravitational settling, wet scavenging by clouds and precipitation, and even their chemical processes (Textor et al., 2007; Wu et al., 2020). For example, the sulfate(Fig. 12a) uncertainty is generally larger over eastern China and the Sichuan Basin than over northern India,which probably results from different gas-phase and aqueous-phase conversion from SO2except for the above reasons. Large uncertainty over the Sichuan Basin is also caused by unique topography (Liu et al., 2021). The BC(Fig. 12c) uncertainty is relatively weaker as the results are mainly determined by the prescribed anthropogenic emissions. For OA (Fig. 12b), the concentration differences also may be caused by the way that models represent various natural biogenic VOC (SOA precursors)emissions.

Natural aerosols are important sources of uncertainty in PM2.5simulation among ESMs. The PM2.5DU uncertainty prevails over northwestern China and Mongolia,the Mongolian Plateau, and northwestern Indian Peninsula (Fig. 13a). There are many reasons for the significant model-spread in dust simulations. In addition to the effects of dynamic transport, and wet and dry depositions, large model-spread is mainly caused by the difference in driving mechanisms of dust emissions that depend on the meteorological drivers (winds and precipitation), especially in East China and South Asia, associated with large-scale monsoonal circulations (Wilcox et al., 2020; Zhao et al., 2022), the land surface conditions(Aryal and Evans, 2021), and the representation of aerosol size distributions (Zhao et al., 2022). And the model complexities also have the influence on dust concentrations (Zhao et al., 2022). As for sea salt aerosols (Fig.13b), it has lower concentrations than other species, and its spreads among models are less than 1 μg m?3over land. Sea salt emissions are mainly determined by near surface wind across the ocean (Wu et al., 2020). It is possible that there is a small model-spread in surface winds across the ocean leading to less spread in sea salt emissions, although inter-model differences in advective transport, and wet or dry deposition will be similar to those for dust (Witek et al., 2007; Wu et al., 2020),which can also affect the simulation of sea salt.

Fig. 7. As in Fig. 5, but for the sulfate.

The Taylor diagram in Fig. 14 statistically examines the spatial distribution as well as the spatial variability of the differences between ESMs and MERRA-2 for main species of PM2.5. The spatial distribution of BC concentrations simulated by ESMs are the best captured with spatial correlation coefficients of 0.9?0.97, followed by sulfate, OA, PM2.5DU, and PM2.5SS. For PM2.5DU, there are large differences between the individual ESMs and MERRA-2, with normalized standard deviations ranging from 0.2 to 3.5 and spatial correlation coefficients from 0.4 to 0.87. The normalized standard deviations of CESM2-WACCM and MRI-ESM2-0 are greater than 2,indicating that the spatial variability of PM2.5DU is largely overestimated in the two models. Although the spatial correlation coefficient of PM2.5SS can be 0.95 or higher, the normalized standard deviations of less than 0.6 in all ESMs, resulting from the overestimation of PM2.5SS in MERRA-2. In general, although there are differences between individual ESMs, the MME can still capture the spatial distributions of five components from PM2.5well compared to MERRA-2. The spatial variations in ESMs are larger than MERRA-2 for OA, BC,and PM2.5DU.

Fig. 8. As in Fig. 5, but for the organic aerosol.

Fig. 9. As in Fig. 5, but for the black carbon.

4.2 Uncertainty in dominant PM2.5 components over different subregions

Each component of PM2.5has different contributions to the PM2.5concentrations in various regions, and the contributions vary between the individual ESMs. Here,we analyzed four regions as illustrated in Fig. 2, Central Asia (CA), East Asia (EA), South Asia (SA), and Southeast Asia (SEA). In the whole Asian region (5°–55°N,70°–140°E), the area-averaged MME PM2.5is smaller than for MERRA-2 (by 3.7 μg m?3, Fig. 15), which is largely attributed to their difference in PM2.5SS. The main PM2.5components in Asia are sulfate and OA, accounting for 28% and 32% of the PM2.5in the MME, respectively. PM2.5DU is the third main PM2.5components in Asia, accounting for 21% of the PM2.5in the MME.The largest model-spread among the five main PM2.5species comes from PM2.5DU (Fig. 16), indicating its largest contribution to the PM2.5uncertainty over Asia.

Fig. 10. As in Fig. 5, but for the PM2.5 fine particles of dust.

The proportion of each PM2.5component has large regional characteristics (Fig. 16). PM2.5DU plays a dominant role over central Asia, accounting for 70% of the PM2.5concentration. There are also considerable differences in PM2.5DU model results over central Asia and the uncertainty range is almost 25 μg m?3. In East Asia,sulfate and OA are the main PM2.5species, and the uncertainty is mostly attributed to PM2.5DU and sulfate. In South Asia, the uncertainty ranges are comparable for sulfate, OA, and PM2.5DU. In Southeast Asia, PM2.5SS accounts for 35% of the PM2.5in the MME, and it has the largest contribution to the PM2.5uncertainties. Overall, it appears that the regions of large model diversity are consistent with high concentrations areas for the five components.

Fig. 11. As in Fig. 5, but for the PM2.5 fine particles of sea salt.

5.Uncertainties in evaluating PM2.5 concentrations

The above analyses have shown that surface PM2.5concentrations from ESMs simulations are lower than those from individual observations at CNEMC and EANET sites. One possible reason is the spatial heterogeneity of ground-based observations and the urban effect on PM2.5concentrations. It is noticed that all the CNEMC sites are located in urban area, whereas ESMs simulate average PM2.5concentrations across a coarse model grid larger than 100 km and are hard to identify the differences between urban and suburban areas. Figure 17a shows time series of surface PM2.5concentrations at one city and its neighboring suburban site in Thailand from APAD data (Cohen and Atanacio, 2015).It is clear that the surface PM2.5concentrations at the urban location are evidently higher than those at the neighboring suburban site. The urban site in Thailand is in a residential-university-shopping district containing commercial buildings and small industrial factories. The emissions mainly come from human activities (including automobile exhausts, residential cooking, and heating from buildings). By contrast, the suburban site is surrounded by residential areas with brick-timbered houses,trees, and grass. Urban observatories are more polluted than suburban ones, even when they are geographically close to each other. This is also evident in the two pairs of urban and neighboring suburban sites in China (Figs.17b, c). Differences between downtown and suburban sites in the same city may be higher than 10 μg m?3, and the results in ESMs are closer to those at suburban sites.

Fig. 12. The model-spread of annual mean concentrations (μg m?3)for anthropogenic aerosols during 2005?2020. (a) Sulfate, (b) OA, and(c) BC.

Fig. 13. As in Fig. 12, but for the natural aerosols. (a) PM2.5DU and(b) PM2.5SS.

Another important reason for the uncertainty in evaluation is the method to calculate PM2.5concentrations.Firstly, Eq. (1) used in this study does not include all the aerosol components that constitute PM2.5, such as ammonium and nitrate aerosols, which are generally included in observations but not model derived PM2.5, especially important over eastern China where nitrate aerosols may be responsible for over 20% of PM2.5mass concentrations in winter (Liu et al., 2017). In addition, Eq.(1) assumes fixed percentages of the total mass of dust(10%) and sea salt (25%) aerosols present within the fine size fraction (i.e., less than 2.5 microns in diameter),which are not consistent among ESMs, and also is not suitable for the MERRA-2 data.

Fig. 14. Taylor diagram of the annual mean surface components(sulfate, organic aerosols, black carbon, PM2.5DU, and PM2.5SS) concentrations simulated by the 10 ESMs compared with the MERRA-2 reanalysis data during 2005–2020 in Asia (5°–55°N, 70°–140°E). The radial coordinate shows the standard deviation in the spatial pattern,normalized by the observed standard deviation. The azimuthal variable shows the correlation of the modeled spatial pattern with the observed spatial pattern.

Fig. 15. Histograms of 2005–2020 averaged concentrations (μg m?3)of PM2.5 and their components (sulfate, OA, BC, PM2.5DU, and PM2.5SS) from 10 ESMs, their MME, and MERRA-2 for Asia(5°–55°N, 70°–140°E). The mean value in MME and model diversity for the five main PM2.5 species are 3.5 ± 1.23 μg m?3 for sulfate, 3.98± 0.98 μg m?3 for OA, 0.86 ± 0.15 μg m?3 for BC, 2.59 ± 1.57 μg m?3 for PM2.5DU, and 1.5 ± 0.83 μg m?3 for PM2.5SS.

Fig. 16. Distribution of differences for PM2.5 and their components(sulfate, OA, BC, PM2.5DU, and PM2.5SS) concentrations (μg m?3)from 10 ESMs in Asia and four subregions during 2005–2020. The box plots show the 25th and 75th percentiles as solid boxes, median values as solid lines, dots represent the concentrations from MME, and whiskers extending to the minimum and maximum.

6.Summary

This study uses five main components of aerosols (i.e.,sulfate, organic aerosol, black carbon, dust, and sea salt)that are simulated by 10 CMIP6 ESMs to calculate surface PM2.5concentrations over Asia. Ground-based observation networks as well as the MERRA-2 reanalysis are used to evaluate the ability of current ESMs to simulate PM2.5and its components. In Asia, PM2.5accounts for more than 30% of the total aerosol (including all particle sizes), except for central Asia. The spatial distribution of PM2.5and its main components in the MME are in a good agreement with MERRA-2 and available ground-based observations. High PM2.5concentrations (>40 μg m?3in MERRA-2) are simulated in three regions,eastern China and northern India mainly consisting of anthropogenic aerosols, and northwestern China due to high concentrations of mineral dust. The contribution of each aerosol component to the MME PM2.5across Asia are mainly from sulfate (28%), OA (32%), and PM2.5DU(21%). The proportions of components making up the MME PM2.5are also regionally dependent. PM2.5DU accounts for more than 70% of PM2.5in central Asia and PM2.5SS for about 35% of PM2.5in Southeast Asia in the MME.

Our analysis shows that PM2.5from ESMs is biased low in the comparison with ground-based observations. It may be partly due to the unevenly distributed groundbased observations and the effect of urban areas, as well as the formula used to derive the PM2.5concentrations in this work, which does not consider the contributions of nitrate and ammonium compounds. Compared to the MERRA-2 reanalysis data, the MME underestimates PM2.5concentrations averaged across Asia by about 3.7 μg m?3, which is possibly due to large PM2.5SS overestimation in MERRA-2.

Fig. 17. Time series of surface PM2.5 concentrations (μg m?3) in neighboring city and suburban from APAD, CMA, CNEMC, and MME. Red and blue lines represent observations at urban and suburban sites, respectively. Black lines represent the simulations from MME.

There are large uncertainties in simulations of PM2.5and its components among the 10 ESMs. Inter-model differences in PM2.5are mainly attributed to sulfate and PM2.5DU over East Asia, and PM2.5DU over central Asia. For South Asia, the uncertainty ranges are comparable for sulfate, OA, and PM2.5DU. PM2.5SS has the largest uncertainty range in Southeast Asia. The differences in the simulation of PM2.5and its components amongst the 10 ESMs to a large extent reflect the different algorithms used to prognose aerosol variations in the individual ESMs including the dynamic transport, dry deposition, gravitational settling, wet scavenging, chemical processes, meteorological drivers, land surface conditions, and the representation of aerosol size distributions.

This work is the first to highlight ESM model biases in the simulation of PM2.5concentrations across Asia using observations and a reanalysis dataset. Analyzing the individual aerosol components highlights the potential improvements to ESMs and the certain aspects of their individual aerosol schemes to target. It is noted that the ground-based observations used in this work are relatively sparse. The regional feature for PM2.5and its components in ESMs still needs further investigations using more data with high spatial and time resolutions that retrieved from satellite observations (Wei et al., 2020; Yan et al., 2020, 2021) in the future.

Acknowledgments.We would like to thank Meteorological Observation Center, China Meteorological Administration for providing surface PM2.5data at atmospheric background stations.