Simulating Aerosol Optical Depth and Direct Radiative Effects over the Tibetan Plateau with a High-Resolution CAS FGOALS-f3 Model

The Tibetan Plateau (TP), known as the Third Pole, is the highest and largest plateau on Earth, with an area of approximately 2.5 × 10⁶ km² (Qiu, 2008). The TP is one of the most important stores of ice, and the melting of glaciers over the TP has been providing abundant water resources to the Yangtze River, Yellow River, Indus River, and Ganges River (Singh and Bengtsson, 2004; Xu et al., 2008; Immerzeel et al., 2010). The TP is not only an obstacle to airflow but also an enormous heat source and sink affecting the large-scale atmosphere circulation (Wu et al., 2017; Zhu et al., 2018). Under the background of global warming, the climatic conditions over the TP have drastically changed due to pronounced increases in human activity (Hansen and Nazarenko, 2004; Flanner et al., 2007). According to observed temperature records and reanalysis datasets, the TP has been warming faster than the global average over the last 60 years (Duan and Wu, 2006), revealing that the TP is one of the most sensitive indicators of climate change (Wan, 2018).

The TP has been surrounded by pollutants from various sources. Aerosols are among the most uncertain factors affecting climate change over the TP due to their heterogeneous distributions and complicated interactions with clouds and radiation (Ghan and Easter, 2006; Zhang et al., 2012a). Aerosol particles significantly influence the radiative budget by scattering and absorbing radiation (Charlson et al., 1992; Huang et al., 2010; Che et al., 2018; Zhao et al., 2018, Letu et al., 2022). Much work has been done on the aerosol direct radiative effect (DRE) in global areas. Zhao et al. (2015) suggested that the asymmetric cooling effect of dust between the northern hemisphere (NH) and southern hemisphere (SH) leads to a more severe reduction in evaporation over the low latitudes of the NH compared with that in the SH. The aerosol over the TP is also modulated by atmospheric variations and, in turn, affects atmospheric circulation through both direct and indirect radiative effects (Yang et al., 2018). In addition, the aerosols, including black carbon (BC) and dust deposited in the ice and snow coverage area of TP, significantly change the surface albedo affecting the reflection of solar radiation (Warren and Wiscombe, 1980; Ming et al., 2009; Niu et al., 2020). Recent studies have found that the dust and BC over the TP may accelerate snow and ice melting by reducing the snow albedo and generating surface radiative flux changes ranging from 5 W m^–2 to 25 W m^–2 during the springtime (Qian et al., 2011). The BC DRE could increase the mid-tropospheric temperature and change the precipitation due to the large geographical extent of the forcing over the TP, whereas the BC snow albedo effect is highly localized over the snow cover region (Usha et al., 2020). Whether absorbing aerosols exert a net cooling or heating effect at the top of the atmosphere (TOA) depends largely on whether they have a low or high albedo surface (Haywood et al., 1995; Haywood and Boucher, 2000; Doherty et al., 2022). There is still great uncertainty in the aerosol DRE over the TP. Using a chemical transport model (CTM), Kopacz et al. (2011) estimated BC DRE at the TOA to be 0.2–1.7 W m^–2 at the five Tibetan sites. Wang et al. (2014) further obtained a much smaller BC DRE with values lower than 0.5 W m^–2 at the TOA using the same CTM and updates. He et al. (2014) estimated the annual BC direct radiative forcing (DRF) at the top of the atmosphere to be 2.3 W m^–2 with the uncertainties of –70%–85% in the TP. Sadavarte et al. (2016) used simulated aerosol fields from the Weather Research and Forecasting (WRF) model coupled with the chemical transport model STEM and the OPAC-SBDART radiation transfer model to calculate the aerosol DRF and heating rate and found significant anthropogenic contribution to the atmospheric forcing and heating rate over the remote TP.

Numerical models are important tools for studying the climatic effects of aerosol particles. Researchers have found that the simulations tend to be closer to the observations as the model's horizontal resolution increases (Qian et al., 2010; Ma et al., 2014; Sato et al., 2016). Gan et al. (2016) illustrated that finer horizontal resolution is beneficial for local air quality since it can provide information on local gradients. Higher resolution does not always favor all physical processes and parameterizations in the model, nor can it always eliminate the proposed systematic errors in the model (Demory et al., 2014). Varghese et al. (2011) explored aerosol surface concentrations in a regional climate-chemistry-aerosol model and found no significant improvement with increased resolution; however, the precipitation appears to improve. Despite these issues, improving model resolution is generally considered an important way to improve model performance. In particular, the complex topography over the TP further highlights the importance of high-resolution global climate models (GCMs), as the coarser resolution of GCMs makes it difficult to capture the surface details of the TP. Li et al. (2015) demonstrated that the low-resolution model inadequately reproduces the narrow large-rainfall belt along the southern edge of the TP and that the spatial distribution of precipitation over the TP significantly improves with increased resolution. Na et al. (2021) indicated that the 14-km Nonhydrostatic ICosahedral Atmospheric Model well-reproduces the historical precipitation spatial pattern, seasonal cycle, and the extreme precipitation over the southern slope of TP while, at the same time, overestimating the precipitation amount by ~35%. Zhang et al. (2020) used the WRF-Chem model and found that the 4-km complex topography results in stronger BC transport across the Himalayas due to the strengthened efficiency of near-surface transport towards the TP in a high-resolution model. However, to our knowledge, aerosols and their DRE over the TP have not been studied with high-resolution GCMs.

To examine the impacts of the high-resolution global climate model on aerosols and their DRE over the TP, this study conducts two experiments with CAS FGOALS-f3 that has recently been coupled online with an aerosol module (Wang et al., 2020). The model coupled aerosol module is suitable for simulations at hydrostatic and non-hydrostatic scales and thus can be used for investigating the influence of aerosols from different sources on aerosol surface concentrations and aerosol optical depth (AOD) over the TP (Zhao et al., 2021a). In particular, the AOD and meteorological fields simulated in CAS FGOALS-f3-L in the TP and its surrounding areas have been systematically evaluated using the Moderate Resolution Imaging Spectroradiometer (MODIS) retrieved AOD and the ERA5 reanalysis monthly mean data, respectively. The model can capture the inter-annual and seasonal variation of AOD and meteorological fields. (Zhao et al., 2021a, b). All previous studies lay the foundation for this study.

The organization of the paper is as follows. Section 2 briefly describes the CAS FGOALS-f3 model, the aerosol module, the model configuration, and the experimental design for this study, followed by the description of the data to be evaluated. The evaluation of the numerical experiments at different resolutions and the reasons for AOD differences due to different resolutions over the TP are analyzed in sections 3.1 and 3.2, respectively. The aerosol DRE over the TP simulated at different resolutions is given in section 3.3. The findings are summarized in section 4.

The model used in this study is CAS FGOALS-f3, which is a global climate model (GCM) developed by the State Key Laboratory of Numerical Modeling for Atmospheric Sciences and Geophysical Fluid Dynamics (LASG) at the Institute of Atmospheric Physics (IAP), Chinese Academy of Science (CAS). The Finite-volume Atmospheric Model Version 2.2 (FAMIL2.2) serves as the atmospheric component of CAS FGOALS-f3 (Bao et al., 2010, 2020; He et al., 2019, 2021; Li et al., 2019). The FAMIL2.2 owns the dynamic core of a finite-volume cubed-sphere that covers six tiles on the globe (Lin, 2004; Putman and Lin, 2007; Zhou et al., 2015), and each tile can model air motions with a horizontal resolution ranging from about 200 km (48 grid cells in each tile) to about 6.25 km (1536 grid cells in each tile). The one-moment cloud microphysical parameterization used in this model determines the mass mixing ratio of six categorized hydrometeors, including water vapor, cloud water, cloud ice, rain, snow, and graupel (Lin et al., 1983). The shortwave and longwave radiative transfer scheme is the Rapid Radiative Transfer Method for GCMs (RRTMG) (Clough et al., 2005).

The CAS FGOALS-f3 is coupled to an aerosol module called the Spectral Radiation Transport Model for Aerosol Species (SPRINTARS) (Takemura et al., 2005; Wang et al., 2020). As an online coupled GCM, the meteorological, transport, and aerosol components, including dust, sea salt, BC, organic carbon (OC), and sulfate, are integrated simultaneously. Meteorological fields control the emission, advection, diffusion, and deposition of aerosols, and the forcings from the aerosol fields can, in turn, feedback to the meteorology. The particle radii of soil dust and sea salt in SPRINTARS are divided into 10 bins and 4 bins, respectively, ranging from 0.1 to 10 μm, and they are treated as spherical particles. The emission flux of dust and sea salt are calculated online in this model (Takemura et al., 2000; Gong, 2003; Dai et al., 2018). The emission sources of carbonaceous aerosol and sulfate precursors from anthropogenic sources and biomass burning sources are from the Community Emissions Data System (CEDS) in the Climate Model Inter-comparison Project 6 (CMIP6: https://esgf-node.llnl.gov/search/input4mips/; Van Marle et al., 2017; Hoesly et al., 2018) datasets. The gas-to-particle conversion of terpene is from the MEGANv2.10 biogenic emission inventory (https://eccad3.sedoo.fr/). We first assume an external mixture of most aerosol components. Secondly, we assume that the internal mixture of carbonaceous aerosol of four different types can be used in aerosol transport and radiation processes (Takemura et al., 2009). Mie calculations calculate the aerosol optical properties by summing over all size bins to obtain the single scattering albedo, asymmetry factor, extinction coefficient, and scattering coefficient (Sekiguchi and Nakajima, 2008; Dai et al., 2014).

Two numerical experiments were conducted to investigate whether the impact of model resolution altered the AOD and aerosol DRE. One experiment, called the HRM, has a finer grid spacing of approximately 25 km, and the other experiment, called the LRM, has a coarser grid spacing of about 100 km. The same 32 vertical layers with a top of 2.16 hPa are used in the LRM and HRM. The main difference in experimental settings between HRM and LRM is the time steps required to maintain the stability of the dynamic core. The time steps are set to 15 and 30 minutes, and the radiative module is called every 7.5 and 15 minutes in HRM and LRM, respectively. The terrain output corresponding to the two different resolutions over the TP is shown in Fig. 1. The two experiments are simulated from 2000 to 2014, and the results from the last 10 years are used for analysis.

Figure 1.  Spatial distributions of terrain height from the low-resolution model (LRM: ~100 km) (top row) and high-resolution model (HRM: ~25 km) (bottom row). The white line represents the boundary of the Tibetan Plateau. The red dots represent the seven sites in AERONET, and the red triangle represents the Lhasa site.

Seven AErosol RObotic NETwork (AERONET) sites over the TP provide ground-based observations of AOD. Figure 1b denotes the locations of the seven sites with red dots, and Table 1 lists the location and observational period of the sites in detail. The Muztagh_Ata site is located in the northeastern TP, which is close to the Taklimakan Desert; the Litang site is located in the eastern TP and is vulnerable to pollution; the Langtang site is located along the southern slope of the Himalayas; the NAM_CO, QCMS_CAS, and EVK2-CNR sites are located in the southern TP with elevations above 4000 m; and the Jomsom site is located at the southern TP with an elevation of less than 3000 m. The available AOD observations in AERONET Level 2 are passed to cloud-screening at monthly levels, mainly at 340, 380, 440, 675, 870, and 1020 nm (Smirnov et al., 2002, 2009). The AOD at 550 nm is obtained by linearly interpolating the AERONET-retrieved AODs at 440 and 675 nm, which is used to evaluate the simulated AODs (Cheng et al., 2021). We use the bilinear interpolation method to address the spatial resolution discrepancies between the simulation and AERONET data to interpolate the HRM and LRM data to the AERONET site.

This study obtained the monthly mean surface mass concentrations of dust, BC, OC, and sulfate from 2006 to 2007 at the Lhasa site from Zhang et al. (2012b). Figure 1b denotes the Lhasa site with a red triangle.

The fifth generation of reanalysis product (ERA5) comes from European Centre for Medium-Range Weather Forecasts (ECMWF). The ERA5 is produced based on the ECMWF forecasting model using four-dimensional variational analysis (4D-Var) data assimilation with 137 hybrid levels at the top of 0.01 hPa, which performs well in meteorological variables (Betts et al., 2019; Lei et al., 2020; Yang et al., 2020). The monthly averaged relative humidity at 500 hPa with the resolution of 0.25° is used in this study for simulations comparison. Data can be ordered from the website: https://www.ecmwf.int/.

According to the CMIP6 experimental design, we focus on calculating the all-sky DRE for different types of aerosols in this study. The aerosol DRE represents the instantaneous radiative impact of different types of all aerosols. The radiation module is called in every radiation time step to estimate the DRE of different component aerosols for the present time in HRM and LRM. The aerosol all-sky DRE is the difference between the solar radiation flux with and without aerosols at the TOA or surface. The shortwave DREs can be decomposed as follows:

F_all-sky (aerosol) represents the shortwave radiation flux with aerosols at the TOA or surface, and F_all-sky (no aerosol) represents the shortwave radiation flux without aerosols at the TOA or surface.

Figure 2 shows the climatology of simulated AOD for 2005–14 in different resolution experiments, within which the TP and neighboring dust emissions and anthropogenic emission sources lie. Focusing on the annual AOD averaged in the TP, the simulated AOD in HRM and LRM are 0.197 and 0.212, respectively. In general, both simulations reproduce the overall spatial distribution of AOD, with high AOD values (> 0.2) in the northeast TP and low AOD values (< 0.1) in the central and southeast TP, which may be attributed to the terrain features, aerosol transport, and human activity. Fine features of dust AOD patterns are notable, particularly over the northeastern TP and the Himalayas; however, the HRM reveals a much more detailed pattern in the simulation of AOD components.

Figure 2.  Spatial distributions of annual mean aerosol optical depth from the simulations with LRM and HRM for the 2005–14 period. The circles represent site observations from AERONET. The value in the top-right of each subgraph is the average within the region of the black line.

The black circles in Fig. 2 represent the spatial distributions of the mean AOD from the available AERONET datasets. These circles can only represent the averaged distribution of available AERONET observation data.

Figures 3 and 4 display the comparisons of the dust (DU), carbonaceous (CA), and sulfate (SU) AOD from the simulations in HRM and LRM. The value of sea salt (SS) over the TP can be ignored. The dominant aerosol is dust over the TP; in both the HRM and LRM simulations, dust accounted for more than 70% of the total aerosol. Compared to the dust in the LRM simulation, the dust AOD pattern is better resolved in the HRM, especially in the Taklimakan Desert. Figure S1 in the Electronic Supplementary Material (ESM) plots the comparison of the model results with the MODIS and Multi-angle Imaging SpectroRadiometer (MISR) observations. The AOD in the HRM and LRM runs are projected onto the MODIS and MISR grids with the local area averaging interpolation method. The AOD in LRM was underestimated with a deviation greater than –0.3 in the Taklimakan Desert in spring and overestimated in the east of the TP in four seasons compared with MODIS and MISR. The increase of dust on the AOD simulation in spring and the decrease of anthropogenic aerosol AOD simulation in the HRM effectively improve the simulation over the TP. However, the AOD in the HRM was overestimated in the eastern TP in winter, which is due to the higher dust emissions simulated in the HRM than that in LRM in winter. The seasonal mean simulated AOD in HRM is highest in spring in the TP (0.24), while the value in LRM is highest in summer (0.31). As the dominant aerosol, dust presents the largest AOD with a value of 0.17 in spring in the HRM, which is mainly responsible for the seasonal variation of AOD over the TP. The high carbonaceous AOD exists near the southern Himalayas, reaching 0.2, which is mainly contributed by the biomass burning emission, while the value reduces significantly to less than 0.02 over the TP. The high sulfate AOD exists near the eastern TP. The simulated values and the relative contribution of carbonaceous and sulfate AODs to the total AOD in the HRM are generally lower than those in LRM, mainly due to the changes in the meteorological field variables in different resolutions models. The reason will be explained in detail in section 3.2.

In addition, the comparison of the modeled AOD against the temporal and spatial collocated AEROENT AOD is shown in the scatter diagram in Figs. 4c and 4d. Considering the AERONET AOD is subject to errors and uncertainties, we use the mean absolute difference (DIFF) to quantify the difference between modeled and measured AOD. The simulated AOD at the Muztagh_Ata site by the HRM produces a clearer gradient from the Taklimakan Desert to the TP. Over the eastern TP, the AOD at Litang simulated by HRM is obviously closer to the observations than that simulated by the LRM. The NAM_CO, QOMS_CAS, and EVK2-CNR sites are nearby, located in the southern TP with elevations above 4000 m. The observations for these three sites are always less than 0.1 for most of the full year, and the positive DIFF against AERONET data of the AOD decreased in the HRM, although the HRM and LRM both overestimate the AOD. The Langtang station, located on the southern slope of the TP, is mainly under the influence of South Asian pollutants. The simulated AOD in the HRM is in better agreement with the observations than that of the LRM at the seven TP sites. The statistical metrics in Figs. 4c and 4d are calculated based on the values using temporal and spatial collocated data. The AOD DIFF is reduced from 0.15 in the LRM simulation to 0.07 in the HRM simulation, and the DIFF in the HRM decreases by 53%. The root mean square error (RMSE) is also greatly reduced from 0.21 in the LRM to 0.09 in the HRM, and the correlation coefficient (CORR) between the simulations and observation increases from 0.24 (LRM) to 0.46 (HRM), which represents an improvement of 92%. The CORR of AOD between the HRM and AERONET is not above 0.5 because (1) the dust emission coefficients without nudging the wind field in the CAS FGOALS-f3 model; (2) the terrain of the Qinghai Tibet Plateau is complex, and the parameterizations in the global climate models may be quite different from that in the TP; (3) there are some uncertainties in the observations. Despite improvements of the spatial distribution of AOD in high-resolution simulations over low-resolution simulations, there is still an overestimation compared with the observed value. One possible reason for this overestimation is high relative humidity leading to excessive moisture absorption growth of hydrophilic aerosol or an anomalously high-wind field leading to the excessive generation of dust aerosol. The other possible reason is too much transportation from the outside of the TP and the overestimation of the mass extinction coefficient.

Figure 3.  The annually-averaged aerosol optical depth of dust, carbonaceous, and sulfate from the simulations with HRM and LRM during the 2005–14 period. The average of each aerosol component over the Tibetan Plateau is labeled at the top-right of each subgraph. DU: dust; CA: carbonaceous; SU: sulfate.

Figure 4.  Comparison of the different component seasonal mean AODs over the Tibetan Plateau between the HRM (a) and LRM (b). The scatter diagrams represent the AERONET AOD against the HRM AOD (c) and LRM AOD (d).

Observations of different components aerosol concentrations at Lhasa in 2006 and 2007 were collected from Zhang et al. (2012b) and are used to evaluate the model results (Fig. 5). Although overestimations of AOD presented themselves in modeling results, underestimation of aerosol components concentrations exists in the HRM and LRM runs. The aerosol surface concentrations biases are reduced from –10.21 μg m^–3 in the LRM simulation to –6.39 μg m^–3 in the HRM simulation. The BC and OC surface concentrations in the HRM are slightly closer to the 1:1 line compared to observations than that simulated in the LRM. The underestimation of BC and OC surface concentrations indicates that the corresponding aerosol mass extinction coefficient in the model needs to be lowered to achieve better agreement with AERONET observations (Jacobson, 2001; Schulz et al., 2006). In addition, the HRM successfully reproduces the temporal and magnitude variation of dust concentration in April 2007 and June 2007 (Fig. 5c). However, the LRM shows underestimations during all periods; however, the HRM shows overestimations during May 2007 but underestimations from July 2007 to November 2007. The bias of surface wind speed, snow amount, and the uncertainties within the dust emission scheme over TP may contribute to such deviations in modeling results.

Figure 5.  Monthly aerosol surface mass concentrations from available measurements and simulations at the Lhasa site. Scatterplot of observations vs. HRM (a) and LRM (b). The time series of dust surface mass concentrations simulated in HRM and LRM are compared with the observations from Apr. 2007 to Nov. 2007. The shaded area represents the standard deviation of the observations.

Each type of aerosol has a unique response to a change in model resolution. For carbonaceous and sulfate hydrophilic aerosols, the highest AOD differences between HRM and LRM occur over the eastern and southern TP, where carbonaceous and sulfate AODs exist in the context of a sharp gradient, which is mainly due to the Himalayas acting as a distinct boundary line for aerosols are more finely delineated in HRM. In general, the magnitudes of sulfate and carbonaceous AOD simulated in HRM are lower than those simulated in LRM. This is attributed to the fact that changes in sulfate and carbonaceous particle radius and Ångström exponent are sensitive to changes in RH. In most current GCMs, the parameterizations of aerosol size distribution and refractive indices based on the empirical function of RH with limited observations are used to determine aerosol mass extinction efficiency according to Mie-theory (Chin et al., 2002; Liu et al., 2007). In the newly coupled aerosol module to CAS FGOALS-f3, RH affects AOD in two ways. When the RH is below 70%, AOD is insensitive to the change of RH, whereas the response of AOD rapidly changes with the increase or decrease of RH when RH is above 70% (refer to Fig. 14 in Wang et al., 2020). Figure 6b illustrates that higher values of RH (above 75%) at 500 hPa simulated in the LRM occur over the western and northern TP. The RH modeled in the HRM is much lower than that simulated in the LRM, and the smaller RH occurs over the central and northeastern TP, which is close to the ERA5 reanalysis data (Fig. 6). Figure S2 in the ESM also shows that the HRM is slightly better than the LRM in simulating RH when compared to observations. The absolute difference in the annual mean RH between HRM and LRM is as high as –8.99 %. One sensitivity experiment without carbonaceous and sulfate emissions in the TP shows that the simulated carbonaceous and sulfate AOD reductions are smaller than 0.01 compared to the control experiment (not shown). The carbonaceous and sulfate surface concentration over the TP is primarily contributed by the transport from the outside of the TP, especially in summer and autumn (Zhao et al., 2021a). For all seasons, the differences of carbonaceous and sulfate AOD are negative between the HRM subtracted LRM over the TP, especially during summer (Fig. 3 and Fig. 4), consistent with the negative RH differences between the HRM and LRM simulations (Fig. 6e). Notably, the growth of hydrophilic aerosols in models is uncertain because the parameterized fitting curve in calculating the aerosol mass extinction coefficient is often obtained from a wide range of observation data (Chan et al., 2006; Sun et al., 2013; Ding et al., 2021). Due to the different aerosol composition and particle properties in various regions, the sampled observations may not be able to fully represent the complexity of atmospheric processes (Titos et al., 2016; Li et al., 2021), especially in the complex topography of the TP.

Figure 6.  Spatial distributions of relative humidity (RH) at 500 hPa (units: %) from the HRM (a), LRM (b), ERA5 dataset (c), and the difference between HRM and LRM (d) during the 2005–14 period. The comparison of the seasonal mean RH at 500 hPa (e) over the Tibetan Plateau among the HRM, LRM, and ERA5.

The dust aerosol has the largest proportion of aerosol mass and optical properties in the TP. Unlike the direct input of carbonaceous and sulfate aerosol from the emission sources in CMIP6, dust emissions are calculated online, which are sensitive to the change in model resolution. In this model, the parameterization of dust emission is referenced by Wang et al. (2020) and Dai et al. (2018), mainly affected by soil type, soil moisture, surface snow amount, and 10-m wind speed. The soil type in this study is the same as that mentioned in Dai et al. (2018), and the bare soil type is considered to be a potential source of dust emissions, which is the reason that high dust AODs occur over the northeastern TP (i.e., Qaidam Basin) in the HRM and LRM. It is also consistent with station observations that dust storm events are occasionally observed over eastern TP (Wu et al., 2018). Figure 3 shows the dust AOD simulated in the HRM is finer and higher than that in the LRM, which is predominantly attributed to the superior ability of the HRM to resolve the small-scale meteorological phenomena contributing to dust emission, especially in the TP with complex topography. When the surface snow amount is less than 1 kg m^–2, the soil is drier than the threshold soil moisture, and the surface wind speed is higher than the threshold wind speed, the dust emission increases in the model.

The wind speed not only affects dust emission but also affects dust transport. As shown in Fig. 7, the difference in the 10-m wind speed between the HRM and LRM is positive over the TP in DJF, which is partly the reason for the higher dust AOD simulated in HRM than that in LRM during winter. The difference in the 500 hPa wind field between the HRM and LRM also illustrates that the wind field modeled in HRM is more beneficial to the dust transportation from the Taklimakan Desert to the TP relative to LRM. Observations have confirmed that the highest dust AOD occurred in spring, and the lowest occurred in autumn over the TP (Cong et al., 2009; Mao et al., 2013). Therefore, the HRM can more reasonably reproduce the seasonal variation characteristics of dust AOD compared to the LRM (Fig. 4).

Figure 7.  Differential spatial distributions of 10-m wind speed (units: m s^–1) between the HRM and LRM in four seasons. The vectors represent the difference in the wind field at 500 hPa.

Changes in aerosol concentration could significantly influence the radiation budget over TP through DREs (Singh et al., 2019). Therefore, to understand the importance of the horizontal resolution change on aerosol DRE, Fig. 8 shows the annually-averaged all-sky shortwave DREs over the TP, at the TOA, for dust, carbonaceous, sulfate, and all aerosols in HRM and LRM. The surface albedo is an important factor affecting the heating or cooling of absorbing aerosol at the TOA (Rahimi et al., 2019). Both simulations estimate positive annual mean aerosol TOA DRE values over the western TP and negative values over the eastern TP. In the HRM, all-sky aerosol DRE at TOA over the TP is –0.76 W m^–2. There is the largest difference in the dust TOA DRE between HRM and LRM (–0.32 W m^–2), whereas the differences in the other aerosol components between the HRM and LRM are within 0.2 W m^–2, which is consistent with the difference of the AOD (Figs. 3 and 8). The dust TOA DRE is largely determined by both the distributions of dust and scene albedo (i.e., snow cover, bright cloud) over the TP. The aerosol vertical distribution and location relative to clouds also affect the DRE. The simulated annual-averaged dust TOA DRE is –0.41 W m^–2 in the HRM, with local distributions exhibiting both positive and negative values. A positive value means the aerosols absorb more solar energy and warm the TP at the TOA. In bright surface areas covered by snow over western TP, dust aerosols produce strong positive DRE up to 3 W m^–2 due to the dust aerosol absorbing more shortwave radiation over the higher surface albedo of snow-covered areas. Compared to dust DRE simulated in the LRM, the simulation in the HRM generates higher positive dust DRE over the western TP and higher negative dust DRE over the eastern TP. Negative dust DREs over the central and eastern TP are better resolved in the HRM, covering most of the TP relative to that simulated in the LRM. The mean seasonal variations of different aerosol components DRE in the HRM and LRM over the TP are shown in Fig. 9 and Table 2. It is found that the dust DRE at TOA significantly shifts from positive in spring to negative in summer in both the HRM and LRM, which is mainly due to the scene critical albedo determined by both white surfaces (e.g., snow, cloud) and overlaid absorbing aerosol. The larger dust AOD and snow-covered areas coexist in spring, resulting in a higher surface albedo and more positive values of DRE.

Figure 8.  Spatial distributions of dust, carbonaceous, sulfate, and all component aerosol direct radiative effect (units: W m^–2) at the top of the atmosphere (TOA) for the 2005–14 period from the simulations in the HRM and LRM.

Figure 9.  Comparisons of the seasonal mean different component aerosol direct radiative effects (units: W m^–2) at the TOA (a–b) and the surface (c–d) over the Tibetan Plateau between the HRM and LRM.

Figure 10.  Same as Fig. 8, but for the aerosol direct radiative effect (units: W m^–2) at the surface.

Both simulations show that the carbonaceous DREs are positive with values of 0.28 W m^–2 and 0.38 W m^–2 in the HRM and LRM, respectively, which is also largely coincident with the spatial distributions of carbonaceous AOD. Because carbonaceous aerosols are considered internal mixtures with BC and organic carbon (OC) in the model, they exert different radiative effects depending on the mass ratio of BC and OC. The carbonaceous DRE over the western TP in HRM, reaching 2 W m^–2, is dominated by BC absorbing effect, while the carbonaceous DRE over the eastern TP in HRM is lower than 0.2 W m^–2, which can be partly attributed to the neutralization of the scattering of OC and the absorption of BC. In addition, the surface albedo in the western TP is higher than that in the eastern TP, which can also strengthen the carbonaceous DRE in the western TP. Seasonally, both simulations illustrate that the carbonaceous TOA DRE is higher in spring and summer and lower in autumn and winter. The seasonal variations are consistent with those from He et al. (2014), although their abovementioned results only accounted for the warming effects of BC. The warming effect of carbonaceous TOA DRE is reduced by 16%–40% from the LRM to the HRM simulation, mainly due to the corresponding reduction of positive RH bias from LRM to the HRM. The sulfate aerosol mainly scatters solar radiation and exerts a cooling effect in the HRM and LRM. A lower sulfate AOD simulated in HRM causes a slightly weaker sulfate DRE relative to that in LRM (Figs. 3 and 8). The magnitude of sulfate DRE at the TOA is most strongly negative during summer (Fig. 9), mainly due to the aerosol transport from the outside of the TP. Overall, the aerosol cooling effect at the TOA is enhanced by 56% over the TP when the horizontal resolution is increased from about 100 km to 25 km, and the dust aerosol dominates the aerosol DRE over the TP.

Considering the aerosol DRE at the surface, all aerosol components exert a cooling effect in the HRM and LRM, as shown in Fig. 10. The HRM-simulated value of the aerosol DRE at the surface over the TP is enhanced by 20% compared to the LRM, which is also dominated by dust aerosol. The response of the carbonaceous and sulfate aerosol DREs at the surface to different spatial resolutions results in a reduction of 22% and 20%, respectively, from the HRM to LRM. However, the proportions of carbonaceous and sulfate DREs at the surface to the total aerosol DRE are only 9% and 7% in HRM, respectively, so the change in the aerosol DRE at the surface is similar to the pattern of the dust DRE change. Seasonally, differing from the aerosol surface DRE simulated by the LRM, the seasonal variation of the aerosol surface DRE simulated by the HRM is larger in spring and smaller in winter, which is mainly consistent with the seasonal variation of the AOD pattern shown in Fig. 4.

The comparisons of both AOD and surface mass concentration from the simulated and observed data are shown above. In addition, the vertical distribution of aerosol can affect the DRE to some extent; therefore, we discuss the aerosol profile in the TP region and compare the differences in the aerosol extinction coefficients (AEC) between the HRM and LRM. As shown in Fig. 11, discrepancies between the HRM and LRM simulations are found in the TP regarding the AEC. In particular, high values above 0.15 km^–1 of AEC occur at approximately 700 hPa between 30°N and 33°N in the TP. Maximum values are simulated close to the surface in the LRM, while the HRM indicates a maximum above the surface. High values of AEC also occur at the same height between 90°E and 95°E in the TP. The dust belt region may heat the air by absorbing solar radiation. The diffusion of aerosols in the HRM is stronger than that in the LRM, and it then causes a higher DRE in the TP. The difference in the DRE at the surface between the HRM and LRM over the northeastern TP could be due to the difference in the dust belt region at approximately 700 hPa. The aerosol extinction coefficients greater than 0.17 km^–1 occur at 650–750 hPa between 102 and 105 °E in the TP, caused by anthropogenic aerosols from frequent human activities. The effects of changing the horizontal resolution appear to significantly change the vertical distribution of aerosols, and the change of vertical distribution has an important impact on resolving mixing and transport processes (Frey et al., 2021).

Figure 11.  Comparisons of the vertical structure of the 550 nm aerosol extinction coefficients (km^–1) over the Tibetan Plateau between the HRM and LRM.

In addition, the location of aerosol relative to the cloud and the model biases in cloud fraction will also affect the DRE (Doherty et al., 2022). The above cloud aerosol, especially absorbing aerosol, may scatter and absorb more shortwave radiation due to the underlying bright cloud. Peers et al. (2016) found that aerosol above clouds in climate models is underestimated due to the strong reflection. In the CAS FGOALS-f3 model, the layer-resolved aerosol (e.g., AOD/extinction, single scattering albedo, asymmetry parameter) and cloud (e.g., cloud water content, cloud fraction) properties are simulated at each time step. The radiative transfer module can be performed using these vertical profiles to calculate the TOA and surface fluxes in the presence of aerosol. Fluxes without aerosol can also be calculated by removing aerosol properties from the profiles. Finally, the DRE is the difference between the fluxes with and without aerosol. The discrepancy in simulated aerosol profiles (Fig. 11) and simulated cloud water content profiles (Fig. S3 in the ESM) influence the DRE estimations from the HRM and LRM. The aerosol DRE also depends on the vertical distribution of aerosol and its location relative to the cloud layer (Oikawa et al., 2013). This hot topic can be discussed in the future.

It is also worth noting that the dust particles are assumed to be spherical in this model. While this simplification may impact the extinction efficiency and sedimentation velocity, it also affects the aerosol burden and optical thickness. However, the simplification was demonstrated to have only a limited influence on the DRE at the TOA (Di Biagio et al., 2020). The difference is less than 5% and 10% between considering a more realistic phase function and spherical dust in the shortwave and longwave ranges, respectively (Bellouin et al., 2004; Colarco et al., 2014). In addition, changing the particle shape from spherical to nonspherical makes little difference to the simulated dust surface concentration; in other words, the effects of dust particle shape on global modeling are negligible (Ginoux, 2003).

Generally speaking, our results demonstrate that a global aerosol coupled climate model with a 25 km resolution is better than that with a 100 km resolution to capture the small-scale features for accurately simulating the aerosol cycle and its associated DRE. However, it is still difficult to know what simulation resolution is sufficient to predict the aerosol cycle and DRE accurately. This difficulty stems from the resolution of a global climate model, such as CAS FGOALS-f3, not having the ability to change arbitrarily, and the required computing resources are quite large for higher-resolution simulations. In this study, the calculation cost of the HRM is approximately 20 times that of the LRM, which is about the upper limit of the affordable computing resources we have.

In this study, two experiments with different model horizontal resolutions are conducted by CAS FGOALS-f3 coupled to an aerosol module to illustrate the impacts of different resolutions (HRM: ~25 km; LRM: ~100 km) on aerosols and their DREs over the TP for the period 2005–14. The AOD and aerosol surface mass concentrations from simulations were analyzed. It was found that the HRM can better reproduce the spatial pattern of AOD over the TP, especially in areas with high gradients. The geographical pattern of AOD along the eastern TP and southern edge of the Himalayas becomes more realistic in the HRM. The bias and RMSE of AOD between the modeling results and AERONET observations improved from 0.15 and 0.21 in LRM to 0.07 and 0.09 in HRM, respectively. Both the HRM and LRM underestimate the carbonaceous and sulfate aerosol surface mass concentrations. Still, they overestimate their AOD, which may be partially attributed to the uncertainty of the vertical aerosol distribution and the parameterization of mass conversion into an extinction coefficient in the TP region. In addition, the HRM reproduces the temporal and magnitude variations of dust surface mass concentrations from April 2007 to November 2007 reasonably well.

The AOD over the TP is sensitive to the changes in the meteorological fields in different resolution models. In short, the RH exerts a significant impact on AOD change for hydrophilic aerosols, such as carbonaceous and sulfate aerosols. Compared with the LRM, the more reasonable RH was simulated in the HRM, leading to more favorable hydrophilic aerosol simulations in this model. The HRM can also better resolve the small-scale meteorological phenomena contributing to dust emission over the TP and more reasonably reproduce the seasonal variations of the highest dust in spring.

Finally, the aerosol DREs at the TOA and surface over the TP are systematically estimated in the HRM and LRM. The annually-averaged aerosol DREs in the HRM over the TOA and surface are –0.76 W m^–2 and –8.22 W m^–2, respectively. Both simulations illustrate that the dust TOA DRE is positive in spring and negative in the other seasons. The all-sky aerosol cooling effect at the TOA and surface is enhanced by 56% and 20% over the TP when the horizontal resolution is increased from about 100 km to 25 km, respectively, which are dominated by the change of dust aerosol DRE. Considering the anthropogenic aerosols, the warming effect of carbonaceous aerosols and the cooling effect of sulfate at the TOA are weakened by about 25% with the increased resolution, which is consistent with the change of the AOD of the corresponding species.

Acknowledgements. This study is financially supported by the National Natural Science Funds of China (Grant Nos. 41875133, and 91937302), the Strategic Priority Research Program of the Chinese Academy of Sciences (Grant No. XDA2006010302), the Second Tibetan Plateau Scientific Expedition and Research Program (STEP, Grant No. 2019QZKK0206), the Youth Innovation Promotion Association CAS (2020078), and the International Partnership Program of Chinese Academy of Sciences (Grant No. 134111KYSB20200006). We would like to thank the AERONET and ERA5 data used in this study. The AERONET data are available at https://aeronet.gsfc.nasa.gov/cgi-bin/combined_data_access_inv. The ERA5 products are downloaded from https://www.ecmwf.int/.

Electronic supplementary material: Supplementary material is available in the online version of this article at https://doi.org/10.1007/s00376-022-1424-8.