1 Introduction

Short-lived climate pollutants (SLCPs) have been under extensive scrutiny for their roles in perturbing the Earth’s radiation balance (Charlson et al. 1992; Hansen et al. 1997; Haywood and Boucher 2000; Naik et al. 2013; Shindell et al. 2008; Stohl et al. 2015). They include methane, tropospheric ozone, hydrofluorocarbons, and black carbon aerosol, which mainly exert positive radiative forcing on the earth system. Other components of tropospheric aerosols, such as sulfate, nitrate, ammonium, and organic aerosols, which are not classified as SLCPs, also play an essential role in affecting climate by scattering solar radiation. Moreover, tropospheric aerosols act as ice or cloud condensation nuclei and thus change the cloud albedo or lifetime (Lohmann and Feichter 2005; Quaas et al. 2009; Tao et al. 2012), which is referred to as the aerosol-cloud interaction and causes significant uncertainty in future climate projections (Intergovernmental Panel on Climate Change 2013; Shindell et al. 2013).

The 6th Assessment Report of the Intergovernmental Panel on Climate Change (Forster et al. 2021) reported that the effective radiative forcing (radiative forcing with adjustment of atmospheric state by the climate forcing agent) of aerosols for the period of 1750–2019 due to both the aerosol-radiation interaction and the aerosol-cloud interaction is − 1.06 W m−2 (− 1.92 to − 0.21). The ozone radiative forcing for the same period is + 0.47 W m−2 (+ 0.24 to + 0.71). The total radiative forcing of aerosols and ozone is 1.53 W m−2, nearly half of the radiative forcing of the well-mixed greenhouse gases (GHGs), 3.32 W m−2 (+ 2.84 to + 3.79). While the GHGs are chemically less reactive and evenly distributed globally, the SLCPs and other tropospheric aerosols have short lifetimes of a few days or weeks in the atmosphere except for CH4. They are distributed unevenly around the globe, having significant implications for regional climate (Shindell et al. 2013). The complex and nonlinear aerosol-cloud interactions make the estimate of aerosol radiative forcing even more uncertain (Lohmann et al. 2010; Myhre et al. 2013). For instance, the uncertainty of the radiative forcing of SLCPs is 4 to 5 times higher than that of GHGs (Intergovernmental Panel on Climate Change 2013). Hence, it is crucial to better understand the physical and chemical processes involved in the SLCPs and tropospheric aerosol formations and their interactions with the climate system to reduce the uncertainty of the anthropogenic radiative forcing for future climate projections.

Exploring the chemistry-climate interactions can be fulfilled by using a coupled chemistry-climate model, which explicitly simulates the complex atmospheric chemistry processes and their interactions with the climate system. Due to recent advances in the computation power and parallelization technique, many chemistry-climate models have been developed (Morgenstern et al. 2017; Shindell et al. 2013). They have varying complexity in simulating atmospheric chemistry processes, resulting in significant gaps in estimating the radiative forcing of SLCPs (Myhre et al. 2013; Shindell et al. 2013). Several model intercomparison projects, including the Atmospheric Chemistry and Climate Model Intercomparison Project (ACCMIP) (Lamarque et al. 2013), Chemistry-Climate Model Initiative (CCMI) (Morgenstern et al. 2017), and Coupled Model Intercomparison Project Phase 6 (CMIP6) (Eyring et al. 2016), were conducted to divulge associated uncertainties with coupled chemistry-climate simulations of participating models. The model intercomparisons revealed that the multi-model means are consistent with the present-day observations, but the model spreads appear large, indicating significant model uncertainties in simulating chemistry-climate interactions.

This study introduces a new global atmospheric chemistry-climate model, Global/Regional Integrated Model system Chemistry Climate Model (GRIMs-CCM), developed by coupling the GRIMs general circulation model and GEOS-Chem chemical transport model. The GRIMs is a multiscale seamless atmospheric general circulation model, and its capability of reproducing the climatological mean state of the atmosphere is thoroughly validated by a previous study (Hong et al. 2013). The GEOS-Chem has been widely used and extensively evaluated in the literature, and it is well received that the GEOS-Chem can reproduce the spatial and temporal variability of the SLCPs and other tropospheric aerosols (Bey et al. 2001; Jeong and Park 2017; Jo et al. 2016; Kim et al. 2015; Park et al. 2004).

We first present the general description of the GRIMs-CCM and then evaluate its performance by comparing the simulated chemical species, focusing on aerosol components and ozone, with observations from satellite and surface networks. We also estimate the radiative forcing of aerosols by conducting time-scale experiments using GRIMs-CCM and compare the result with the values from the chemistry-climate models in the literature as an indirect evaluation of the model's capability to simulate the radiative forcing of SLCPs. The model’s performance in reproducing the climatological mean state of the atmosphere has been validated in previous studies (Hong et al. 2013; Jeong et al. 2019).

2 Model Description

2.1 Atmospheric General Circulation Model

The GRIMs is a multiscale atmospheric model system developed for numerical weather prediction and climate studies from regional to global scales (Chang et al. 2013; Hong et al. 2013; Lee et al. 2014). It consists of global/regional atmospheric and ocean models, a single column model, and a data assimilation package. As an atmospheric model component of the GRIMs-CCM, we employ the GRIMs-global model program with a spectral dynamic core using spherical harmonics as a basis function. To avoid the Gibbs phenomenon due to spectral transform, we used a semi-Lagrangian scheme to simulate the advection of hydrometeors (Koo et al. 2022). The default horizontal resolution of the model is T62 spectral truncation, which is equivalent to about a 210 km resolution at the equator and corresponds to 192 × 94 (longitude, latitude) gaussian grids. The default number of vertical layers is 47 in a hybrid sigma-pressure coordinate system, the top of which is 0.01 hPa. The model uses a semi-implicit time integration scheme with a timestep of 20-min. Note that the horizontal and vertical resolutions are customizable by the user to fit the spatial and temporal scope of research interests.

Physics schemes are based on the GRIMs physics package v3.1, thoroughly described and evaluated by Hong et al. (2013). In the GRIMs-CCM, the shortwave and longwave radiation schemes are replaced with the widely used Rapid Radiative Transfer Model for general circulation model (RRTMG) to facilitate the interaction of the radiation and chemical tracers (Clough et al. 2005; Mlawer et al. 1997). The cumulus parameterization scheme is changed to the simplified Arakawa-Schubert scheme developed for the Korean Integrated Model (Han et al. 2020; Hong et al. 2018), which is an updated version of the convection scheme used in the National Centers for Environmental Prediction (NCEP) Global Forecast System (Han and Pan 2011), to improve the sub-grid scale convective transport of chemical tracers. All the references for individual schemes used in the GRIMs-CCM are summarized in Table 1.

Table 1 Overview of GRIMs-CCM model components and their references
Full size table

2.2 Chemistry Modules

The chemistry modules are obtained from the GEOS-Chem v9-01–02 (http://www.geos-chem.org) with several modifications in the source code structure. The GEOS-Chem is an offline 3-dimensional global chemical transport model driven by archived assimilated meteorological data from the Goddard Earth Observation System (GEOS) of the NASA Global Modeling and Assimilation Office (GMAO). The source codes of the GEOS-Chem are modified as subordinate modules of the GRIMs, and the chemistry modules are then driven by simulated meteorology every timestep. The model calculates the emission, deposition, chemistry, and transport processes of 43 chemical tracers, including NOX, OX, CO, non-methane volatile organic compounds (NMVOCs), and aerosol species (Table 2). Some details are described below.

Table 2 List of chemical tracers in the chemistry module
Full size table

2.2.1 Emissions

The global anthropogenic emissions of CO, NOX, and SO2 are taken from the Emission Database for Global Atmospheric Research (EDGAR) global emission inventory (Olivier et al. 1996). If up-to-dated regional anthropogenic emission inventories are available, they are superseded regionally over EDGAR emissions. They include the data from the European Monitoring and Evaluation Programme (EMEP) for Europe, the Big Bend Regional Aerosol and Visibility Observational Study (BRAVO) for Mexico, the Intercontinental Chemical Transport Experiment-Phase B (INTEX-B) for East Asia, the Criteria Air Contaminants (CAC) for Canada, and the National Emission Inventory 2005 (NEI2005) for the United States (Kuhns et al. 2005; van Donkelaar et al. 2008; Zhang et al. 2009).

Anthropogenic emissions of NMVOCs are obtained from the Reanalysis of the Tropospheric Chemical Composition over the past 40 years (RETRO) global emission inventory (Pulles et al. 2007). The aircraft NOX emission is taken from the Aviation Emissions Inventory Code (AEIC) inventory (Stettler et al. 2011), while the soil NOx emissions are calculated following the algorithm of Yienger and Levy (1995). The lightning NOx emissions are calculated based on the temperature profile and the depth of convective clouds from the GCM, as described in Murray et al. (2012). The volcanic SO2 emissions from the eruptive and non-eruptive volcanoes are taken from the Aerosol Comparisons between Observations and Models (AEROCOM) database. The emissions from biomass burning are obtained from the Global Fire Emissions Database version 3 (GFED3) monthly database (van der Werf et al. 2010). The biogenic emissions of NMVOCs, including isoprene and monoterpenes, are calculated from the Model of Emissions of Gases and Aerosols from Nature version 2.1 (MEGANv2.1) using meteorological variables from the online coupled GCM (Guenther et al. 2012). The anthropogenic emissions of primary carbonaceous aerosols are taken from Bond et al. (2007). The natural emission of soil dust aerosols is calculated using either the Goddard Chemistry Aerosol Radiation and Transport (GOCART) scheme (Ginoux et al. 2004) or the Dust Entrainment and Deposition (DEAD) scheme (Zender et al. 2003), as described in Fairlie et al. (2007). The emission of sea salt aerosols from the ocean is parameterized as a function of surface wind speed following the algorithm of Monahan et al. (1986).

2.2.2 Deposition

The removal processes of chemical tracers by deposition include dry deposition in the planetary boundary layer and wet deposition by precipitation. The dry deposition is calculated based on a resistance-in-series model (Wesely 1989) with an updated surface resistance for aerosol particles (Zhang et al. 2001). The dry deposition velocity of aerosol particles over snow or ice is set to 0.03 cm s-1 (Fisher et al. 2011). The gravitational settling of coarse-mode sea salt aerosols (Alexander et al. 2005) and soil dust aerosols (Fairlie et al. 2007) are also considered in the model. The wet deposition of water-soluble chemical tracers includes in-cloud rainout and below-cloud washout due to large-scale precipitation and scavenging of soluble tracers due to convective updrafts (Liu et al. 2001).

2.2.3 Chemistry

The model calculates 111 chemical species concentrations in every grid cell in the troposphere. The chemical mechanism includes coupled oxidants (OX-NOX-NMVOCs)-aerosol chemistry in the troposphere, with 285 gas-phase kinetic reactions, 51 photolysis reactions, and 4 heterogeneous reactions. The reaction coefficients for gas-phase kinetic reactions are calculated in every grid cell using local pressure and temperature from the atmospheric model. The photolysis rate is calculated using the FAST-J scheme (Wild et al. 2000) by assuming an approximate random cloud overlap (Liu et al. 2006). The reactive uptake coefficients for the heterogeneous reactions are taken from Jacob (2000) except for N2O5 hydrolysis, the coefficient of which is from Evans and Jacob (2005). The absorption cross-sections and surface areas of aerosols used for the photolysis and heterogeneous reactions are calculated online using simulated aerosol concentrations assuming the log-normal size distribution. As a default, the Kinetic Preprocessor (KPP) chemistry solver solves the mass balance equations of the reactions (Damian et al. 2002). The SMVGEAR II chemistry solver is an alternative for solving the equation with relatively slow computation speed but high accuracy (Jacobson 1995).

In the stratosphere, a simple chemistry process is utilized. Stratospheric ozone is calculated using a simple linearized ozone parameterization (LINOZ) from McLinden et al. (2000). The LINOZ is a first-order Taylor expansion of local ozone due to its concentration, temperature, and overhead column ozone. The net production rates of stratospheric NOX and HNO3 are prescribed as climatological mean values. The loss of NMVOCs due to OH radical in the stratosphere is calculated using the climatological stratospheric OH concentration. The tropopause height is calculated following the method in Reichler et al. (2003).

The simulation of inorganic sulfate-nitrate-ammonium (SNA) aerosols is thoroughly described in Park et al. (2004). The sulfur oxidation chemistry includes the gas-phase oxidation of SO2 by OH radical and in-cloud aqueous-phase oxidation of SO2 by H2O2 and O3 to produce sulfate aerosol. The nitrate and ammonium aerosol concentrations are determined by calculating thermodynamic equilibrium using the ISORROPIA II module (Fountoukis and Nenes 2007). The simulation of carbonaceous aerosols is as described in Park et al. (2003). The carbonaceous aerosols include hydrophilic and hydrophobic black carbon (BC) and organic carbon (OC). It is assumed that 20% of primary BC emission and 50% of primary OC emission are emitted as hydrophilic, whereas the rest of the emission is emitted as hydrophobic. The hydrophobic carbonaceous aerosols are converted to hydrophilic aerosols with an e-folding time of 1.15 days. The secondary organic aerosols (SOA) from biogenic sources are entirely hydrophilic, assuming 10% yields of SOA from total monoterpene emissions. The simulation of dust aerosols is described in Fairlie et al. (2007), and Alexander et al. (2005) describe the simulation of sea salt aerosols in detail.

2.2.4 Transport

The GEOS-Chem uses TPCORE as an advection algorithm for the chemical tracers (Lin and Rood 1996). In GRIMs-CCM, TPCORE is replaced with the semi-Lagrangian scheme used for the advection of hydrometeors in the GCM for the dynamical consistency. The vertical transport of chemical tracers due to sub-grid convective updrafts and associated horizontal advection of lifted chemical tracers can be a source of large uncertainty in the model (Pouyaei et al. 2021). GRIMs-CCM accounts for updraft, downdraft, entrainment, and detrainment processes of cumulus convection to better represent sub-grid scale convective transport by implementing a scale-aware and physics-based cumulus scheme (Han et al. 2020). For the turbulent mixing of chemical tracers in the boundary layer, either a full boundary layer mixing scheme (TURBDAY) or a non-local boundary layer mixing scheme (VDIFF) (Lin and McElroy 2010) can be used for the turbulent mixing of chemical tracers within the boundary layer.

2.3 Coupling Chemical Processes to GRIMs

Figure 1 shows a schematic diagram of individual processes in the GRIMs-CCM, indicating that simulated meteorological variables from the atmospheric model drive a chemistry simulation. The 3-D ozone and aerosol concentrations and their optical properties are used as input for the radiative transfer calculation in the atmospheric model. The chemistry module is called every timestep with updated meteorological variables. The ozone concentration and aerosol optical properties are transferred to the atmospheric model when the radiative transfer calculation is conducted (typically once per hour). Here aerosol optical properties include aerosol optical depth (AOD), single scattering albedo, and asymmetry parameters. The optical properties of individual aerosol species, including sulfate, OC, BC, sea salt, and soil dust aerosols, are calculated with the MIE theory as a function of 14 shortwave wavelength bands and 16 longwave wavelength bands of the RRTMG (Mishchenko et al. 2002). The dust aerosol size distribution is assumed to follow the gamma distribution, while that of the other aerosol species is assumed to follow the log-normal distribution. The parameters that determine the size distributions' shape are mainly taken from the Optical Properties of Aerosols and Clouds (OPAC) (Hess et al. 1998), as summarized in Table 3. The refractive indices of dust aerosols are adopted from Sinyuk et al. (2003) and others from Hess et al. (1998). The aerosol hygroscopic growth factor as a function of the relative humidity is obtained from Chin et al. (2002) for BC and Hess et al. (1998) for the other aerosols.

Fig. 1
figure 1

Schematic diagram of the structure of GRIMs-CCM

Full size image
Table 3 Size distribution parameters of the aerosol species (RH = 0)
Full size table

3 Model Evaluation

We evaluate the GRIMs-CCM by conducting a 1-year simulation for 2005 after a 1-year spin-up simulation from 2004. The default emission inventories included in the current version of GRIMs-CCM provides the emission data for 2005. The NCEP-DOE AMIP-II Reanalysis (R-2) data (Kanamitsu et al. 2002) are used for the initial condition of the atmospheric model, and the sea surface temperature (SST) and sea-ice concentration are forced by the Global sea-Ice and SST (GISST) data (Rayner et al. 1996) every 24 h. For land cover data, the hybrid STATSGO-FAO soil texture, U.S. Geological Survey (USGS) global land use, and green vegetation fractions climatology from National Environmental Satellite, Data, and Information Service (NESDIS) (Gutman and Ignatov 1998) are used. For emissions, the default anthropogenic emission inventories are adopted as described in Section 2.2. A non-local mixing scheme (VDIFF) is used for turbulent mixing of chemical tracers within the planetary boundary layer, while a KPP chemistry solver is used for solving the mass balance equations of the chemistry mechanism.

We additionally conduct a GEOS-Chem simulation for the same period as a reference. The GEOS-Chem simulation is driven by GEOS-5 assimilated meteorological data with 2° × 2.5° horizontal resolutions. The number of vertical layers in the GEOS-Chem simulation is identical to the GRIMs-CCM simulation, and the same anthropogenic emission inventories with identical chemical mechanisms to GRIMs-CCM are used. The significant difference between the two simulations is that the GEOS-Chem simulation is driven by assimilated meteorological data, i.e., offline run, while the GRIMs-CCM simulation is driven by the meteorological variables calculated from the online atmospheric model. Their difference mainly indicates the importance of meteorological fields and interactive coupling.

The natural emissions of soil dust and sea salt aerosols are highly sensitive to the surface wind speed, so the natural emissions from the two models are likely to be different as the two models use different meteorology. Thus, scale factors are applied to the soil dust and sea salt emissions for the GRIMs-CCM to assure the two models have the same annual global emissions. As the model performance of reproducing the climatological mean state of the atmosphere is already validated in previous studies (Hong et al. 2013; Jeong et al. 2019), we focus on the evaluation of the chemistry module by comparing simulated AOD, ozone, and surface particulate matters from the GRIMs-CCM and the GEOS-Chem with the satellite or in-situ measurements below.

3.1 Aerosol Optical Depth

We first evaluate the 550 nm AOD in terms of global distribution compared to the satellite measurement. We use the standard product of Terra/Aqua Moderate Resolution Imaging Spectroradiometer (MODIS) level 3 monthly 1° × 1° gridded data (MOD08_M3, MYD08_M3) (Levy et al. 2013; Remer et al. 2005) together with the MODIS AOD retrieved from Deep Blue algorithm (Hsu et al. 2006; Sayer et al. 2013) for infilling missing data over the bright surface, which are averaged from 2004 to 2006.

Figure 2 shows the spatial distribution of annual-mean AOD from the observation, GRIMs-CCM, and GEOS-Chem. The MODIS observation shows a high amount of 550 nm AOD over East Asia, northern Africa, and the southern boundary of Tibet. The spatial distribution of AOD is well captured in the GRIMs-CCM, including high AOD over northern Africa due to dust aerosols from the Sahara Desert and an elevated AOD band over the Southern Ocean due to sea salt aerosols. The GRIMs-CCM also captures the high AOD due to anthropogenic aerosols over East Asia and Indo-Gangetic Plain. However, the high AODs over Amazon and Central Africa in the satellite measurements are not captured, indicating that the model underestimates primary carbonaceous aerosols from the biomass burning and the secondary organic aerosols from the biogenic VOC emissions over that region.

Fig. 2
figure 2

Annual mean aerosol optical depth at 550 nm retrieved from the (top) MODIS, and simulated from the (middle) GRIMs-CCM and (bottom) GEOS-Chem. The global mean values are shown in the parenthesis in the upper right corner. Gray-shaded indicates the missing data in the MODIS observation

Full size image

The zonally averaged AOD of different aerosol species is shown in Fig. 3. The observation shows the maximum AOD near 30°N and a secondary peak in the 30°S–60°S latitude band. Both GRIMs-CCM and GEOS-Chem well reproduce the distribution of AOD. In the Southern Hemisphere, the sea salt AOD is responsible for the secondary peak in the 30°S–60°S latitude band, while both dust AOD (0°N-30°N) and sulfate AOD (30°N-60°N) are mainly contributing to high AOD in the Northern Hemisphere.

Fig. 3
figure 3

Zonally averaged annual mean aerosol optical depth for each aerosol species simulated by the (left) GRIMs-CCM and (right) GEOS-Chem. The MODIS observation is shown in black dotted line as a reference. The total AOD (black) is the sum of dust AOD (red), sulfate AOD (green), carbonaceous AOD (sky blue), and sea salt AOD (orange)

Full size image

The correlation coefficient and relative bias of the simulated AOD are quantified at 1° horizontal resolution against the satellite measurements (Table 4). The correlation coefficient of annual-mean AOD with the observation is 0.65 for the GRIMs-CCM and 0.56 for the GEOS-Chem, which are comparable with those of the ACCMIP models (0.46 to 0.64) reported in Shindell et al. (2013). This indicates that the GRIMs-CCM reasonably reproduces the observed spatial variability of AOD. The bias is − 19% for the GRIMs-CCM and − 11% for the GEOS-Chem, which are also in the range of the biases from the ACCMIP models (− 28% to + 54%). Despite using the same emission inventories and chemical mechanisms, the GRIMs-CCM shows lower AOD than the GEOS-Chem. Table 5 shows aerosol budget analysis of each model, indicating that the GRIMs-CCM has shorter lifetimes of aerosols than the GEOS-Chem due to larger wet and dry deposition losses in the GRIMs-CCM. For example, the GRIMs-CCM calculates higher loss frequencies of aerosols due to wet deposition (+ 32% for sulfate, +6% for soil dust, +14% for sea salt) and dry deposition (+ 35% for sulfate, + 33% for soil dust, + 67% for sea salt) than the GEOS-Chem, implying faster removal of aerosols. The faster deposition in the GRIMs-CCM is likely due to strong cumulus convection and surface friction velocity in the model, which increases the chance of aerosol scavenging in the convective updrafts and the chance of particles touching down the surface, resulting in a higher deposition velocity in the model.

Table 4 Statistics for the comparison of simulated AOD with the satellite measurement
Full size table
Table 5 Comparison of loss processes for sulfate, dust, and sea salt aerosols in the GRIMs-CCM and the GEOS-Chem
Full size table

3.2 Ozone

We compare the total column ozone from the model simulations with the satellite measurements. The Aura OMI TOMS-Like Ozone level 3 daily 1° × 1° gridded data (OMTO3d) (Balis et al. 2007; Veefkind et al. 2006) is used as an observation for 2005. Figure 4 shows the seasonal variation of zonally averaged total column ozone from the observation, the GRIMs-CCM, and the GEOS-Chem. In the observation, the higher amount of total column ozone is found in high latitudes, indicative of the ozone transport from the tropics to high latitudes through the Brewer-Dobson circulation. The enhancement of total column ozone in the mid-latitudes is shown during the spring season of each hemisphere. The ozone hole in the polar cap of the Southern Hemisphere is also evident in spring (blue in Fig. 4 left). These seasonal and latitudinal variations are qualitatively reproduced by the GRIMs-CCM and the GEOS-Chem, with a significant underestimation of the Antarctic ozone hole. The LINOZ scheme (McLinden et al. 2000), used to simulate stratospheric ozone in the GRIMs-CCM and the GEOS-Chem, only considers the local tendency as a function of local ozone mixing ratio, temperature, and overhead column ozone. It misses the complex halogen chemistry responsible for the Antarctic ozone hole. Likewise, it is also difficult to simulate the historical trend of total column ozone due to the changes in ozone-depleting substances.

Fig. 4
figure 4

The seasonal variation of zonally averaged total column ozone (in Dobson unit) obtained from the (left) satellite measurement, and simulated from the (middle) GRIMs-CCM and (right) GEOS-Chem. The model data are integrated from the surface to the model top (0.01 hPa). Gray-shaded indicates the missing data in the Aura-OMI observation

Full size image

Compared with the satellite observation and the GEOS-Chem, the GRIMs-CCM shows higher total column ozone in the polar region. Figure 5a shows the differences in zonally averaged annual mean ozone concentrations between the GRIMs-CCM and the GEOS-Chem. It shows that the ozone differences mainly appear in the stratosphere. The GRIMs-CCM shows higher ozone concentrations at 20–40 km altitude in the polar region and lower ozone concentrations at 20–30 km altitude in the tropics than the GEOS-Chem. Figure 5b and c show the differences in temperature and convective mass fluxes between the two models. The GRIMs-CCM shows lower temperature in the stratosphere, which results in higher ozone concentration than the GEOS-Chem as the local ozone tendency negatively correlates with the temperature, as shown in Fig. 5d. Figure 5c shows that the GRIMs-CCM has a stronger convective mass flux in the troposphere. Especially, the stronger convective mass flux near the tropopause can transport the tropospheric air masses into the lower stratosphere, which may lead to a lower ozone concentration in the lower stratosphere.

Fig. 5
figure 5

The difference (GRIMs-CCM minus GEOS-Chem) of zonally averaged annual mean (a) ozone concentration, (b) temperature, and (c) convective mass flux. (d) change of net ozone production in the LINOZ model per local temperature difference from the climatology

Full size image

We also compare the tropospheric ozone concentration at different vertical levels using ozonesonde data from the World Ozone and Ultraviolet Radiation Data Centre (WOUDC). The WOUDC sonde data during 2005 is compared with the ozone concentration from the models. The locations of the WOUDC stations where the sonde data are available for 2005 are shown in Online Resource 1. Figure 6 shows the seasonal cycle of the tropospheric ozone concentration at 750 hPa, 500 hPa, and 250 hPa altitudes, averaged over four different latitude bands, respectively. The model data are sampled at the location of the observational data point before averaging. The GRIMs-CCM reproduces the seasonal cycle of tropospheric ozone concentrations except for the 90°S–30°S latitude band at 500 hPa, showing high correlation coefficients with the observation (0.50 to 0.97). However, the model underestimates the seasonality of 500-hPa ozone in the Southern Hemisphere high latitudes (90°S–30°S). This bias is related to the strong convective mass flux, resulting in weakening stratospheric ozone influxes in the GRIMs-CCM, as shown in Fig. 5c. The GEOS-Chem model shows a high correlation coefficient with the observation at all the latitude bands (0.57–0.94) and generally agrees with the observation, showing the importance of meteorological data for simulating the tropospheric ozone.

Fig. 6
figure 6

Comparison of seasonal cycle of the tropospheric ozone concentration (in ppbv) at three altitude levels (750 hPa, 500 hPa, 250 hPa) and four latitude bands (90°S–30°S, 30°S–EQ, EQ–30°N, 30°N–90°N) following Stevenson et al. (2006) and Young et al. (2013). The model data are sampled at the location of WOUDC station before averaging. The filled circles indicate monthly mean of the observation, and the error bars indicate ± 1 standard deviation. The correlation coefficients are shown in the left or right top of the panels

Full size image

3.3 Surface Particulate Matter (PM)

We evaluate the simulated PM concentrations with observations from the networks, including the Acid Deposition Monitoring Network in East Asia (EANET), the Interagency Monitoring of Protected Visual Environments (IMPROVE) in the United States, and the European Monitoring and Evaluation Programme (EMEP) in Europe. For East Asia, we use PM10 observations for 2005 at 20 EANET sites, measured by automatic monitoring methods, including a β-ray absorption method and a Tapered Element Oscillating Microbalance (TEOM) method. The automatic monitoring data are summed up into monthly data when the data coverage during a given month is more than 50%. The simulated PM10 concentrations are calculated as the sum of their constituents’ concentrations as follows:

$$\begin{array}{c}{PM}_{10}={SO}_{4}^{2-}+{NH}_{4}^{+}+{NO}_{3}^{-}+BC+OC*2.1\\ + Sea Salt (Accum., Coarse)+Soil Dust (Bin1\sim Bin4)\end{array}$$

Figure 7a and b compare the simulated versus observed annual-mean PM10 concentrations in surface air in East Asia. The observed PM10 concentrations are high in the continent and are low in the downwind regions, including the Korean peninsula and Japan. Both the GRIMs-CCM and GEOS-Chem reproduce the observed spatial distribution of PM10 concentrations with high spatial correlation coefficients (0.83 for GRIMs-CCM and 0.73 for GEOS-Chem). However, the normalized mean biases are negative (− 33% for GRIMs-CCM and − 18% for GEOS-Chem), indicating that the models underestimate PM10 concentrations in East Asia. A larger bias in the GRIMs-CCM compared to the GEOS-Chem partly results from soil dust aerosols from the Gobi Desert. The GRIMs-CCM simulates soil dust aerosols from the Gobi Desert, but their amount is much smaller than GEOS-Chem, presumably due to model biases in soil moisture content and surface wind speed.

Fig. 7
figure 7

Comparison of spatial distribution of (a,b) annual mean PM10 mass concentration in East Asia, and (c–f) annual mean PM2.5 mass concentration in (c,d) U.S. and (e,f) Europe. The annual mean concentrations at each site (circle) are shown over the annual mean PM10 or PM2.5 concentrations simulated from the models (shaded)

Full size image

Figure 8a and b show scatter plots of the observed versus simulated seasonal mean surface PM10 concentrations at EANET sites. Both the GRIMs-CCM and GEOS-Chem capture the observed seasonal variation of surface PM10 concentrations in East Asia with high correlation coefficients (0.79 for GRIMs-CCM and 0.67 for GEOS-Chem). However, the regression slope in the GRIMs-CCM is only 0.62 (solid line), again indicating that the model underestimates the observed PM10 concentrations. As shown in Fig. 7a and b, lower soil dust emissions from the Gobi Desert in the GRIMs-CCM likely lead to lower PM10 concentrations compared to the GEOS-Chem.

Fig. 8
figure 8

Scatter plots of seasonal mean (a,b) PM10 concentrations at each EANET site and (c–f) PM2.5 concentrations at (c,d) IMPROVE and (e,f) EMEP sites. The unit is in μg m-3. Each point indicates the seasonal mean PM10 or PM2.5 concentration at each observation site. The regression slope is calculated using reduced major axis method

Full size image

For the United States, we use the PM2.5 observations for 2005 at 178 IMPROVE sites (Fig. 7c and d). Notice that PM2.5 instead of PM10 observations are used here. The daily PM2.5 concentrations at IMPROVE sites are measured by weighing the 25-mm Teflon filter before and after sampling every three days using a microbalance (Malm et al. 1994). The simulated PM2.5 concentrations are calculated similarly to PM10, but the coarse mode sea salt aerosols and soil dust bin 2 to 4 are excluded.

Figure 7c and d show comparisons of annual-mean surface PM2.5 concentrations between the models and the observations in the United States. The observed PM2.5 concentrations are high in the eastern United States and are low in the western United States. Both models reasonably reproduce the observed spatial distribution with high spatial correlation coefficients (0.82 for GRIMs-CCM and 0.81 for GEOS-Chem) and slight overestimations (+ 16% bias in GRIMs-CCM and + 9% bias in GEOS-Chem). The GRIMs-CCM shows high PM2.5 concentrations over the Baja California Peninsula. High concentrations of soil dust aerosols from the Sonoran Desert are simulated from late spring to summer in the GRIMs-CCM. In this season, the high concentrations of dust aerosols affect a large area of Southwestern US in GRIMs-CCM, while the observation and the GEOS-Chem do not show the increase of PM due to dust aerosols.

Figure 8c and d show scatter plots of the observed versus simulated seasonal-mean surface PM2.5 concentrations at IMPROVE sites. Both the GRIMs-CCM and GEOS-Chem reproduce the observed seasonal variation of surface PM2.5 concentrations with high correlation coefficients (0.72 for GRIMs-CCM and 0.68 for GEOS-Chem) and regression slopes close to unity (1.07 for GRIMs-CCM and 1.02 for GEOS-Chem). However, both models show higher concentrations of PM2.5 in December-January–February (DJF) (37% bias in GRIMs-CCM and 58% in GEOS-Chem).

Figure 7e and f compare the simulated versus observed annual-mean surface PM2.5 concentrations at EMEP sites in Europe. The PM2.5 observation data are available at 25 EMEP sites from 11 countries in 2005. In most sites, PM2.5 concentrations are measured by the gravimetric method using high or low volume samplers. An exception is those from Sweden, where the automatic TEOM method is used. The observation shows relatively high PM2.5 concentrations in central Europe and low concentrations in the Iberian and the Scandinavian Peninsula. The simulated PM2.5 concentrations from both models show very similar spatial distribution with a spatial correlation of 0.6. However, both models simulate higher PM2.5 concentrations in central Europe, including Germany and the Czech Republic, than the observation. The normalized mean biases are 64% for GRIMs-CCM and 46% for GEOS-Chem. High concentrations of soil dust aerosols in northern Africa simulated in GRIMs-CCM but not in GEOS-Chem, contribute to the higher bias in GRIMs-CCM.

Figure 8e and f show scatter plots of the observed versus simulated seasonal mean surface PM2.5 concentrations. The models appear to capture the observed seasonal variation of PM2.5 concentration in Europe with correlation coefficients of 0.52 for GRIMs-CCM and 0.56 for GEOS-Chem. The positive bias of simulated PM2.5 concentrations against the observation is the highest in DJF when the normalized mean biases are 89% for GRIMs-CCM and 75% for GEOS-Chem. This bias is mainly due to inorganic nitrate aerosol, often overestimated by several atmospheric chemistry models (Bian et al. 2017; Tuccella et al. 2012; Walker et al. 2012; Zhang et al. 2012).

4 Aerosol Radiative Forcing

The aerosol radiative forcing (RF) is an effective indicator of the impact of aerosols on climate change (Forster et al. 2007; Shindell et al. 2013). The aerosol RF is the imbalance of net radiative flux at the tropopause due to aerosol changes. We estimate the direct aerosol RF due to aerosol changes from the preindustrial era to the present day using 10-year time-slice GRIMs-CCM simulations. Two sets of 10-year model simulations starting from 1850 (preindustrial) and 2000 (present-day) are conducted with the default setting as described in Section 3. For each simulation, the SST, sea-ice concentrations from 1850 to 1859 and from 2000 to 2009 are prescribed. We also consider the temporal change of greenhouse gases by prescribing global mean concentrations of CO2, CH4, N2O, and CFCs from Meinshausen et al. (2011). The default emission databases in the GRIMs-CCM package described in Section 2.2 do not cover the preindustrial era. Thus, we use the emission inventory for ACCMIP (Lamarque et al. 2010), which covers 1850 to 2000, for the anthropogenic and biomass burning emissions. The ACCMIP emission inventory has a time resolution of a decade, so we use the data for 1850 to simulate 1850–1859 and the data for 2000 to simulate 2000–2009. We calculate the radiative flux change due to aerosols by conducting a pair of radiative transfer calculations with and without aerosols for each simulation, and the difference in the flux change at the top of the atmosphere (TOA) due to the aerosols between the two simulations is defined as direct aerosol RF.

Figure 9 shows the simulated aerosol direct RF due to aerosol changes from the preindustrial to the present climate. The simulated global mean aerosol direct RF is − 0.30 W m-2. Values are spatially uneven, showing strong negative RF in East Asia and Europe. Figure 10a–c show that the amount of anthropogenic sulfate-nitrate-ammonium is significantly increased in East Asia and Europe, which explains the strong negative aerosol RF in East Asia and Europe. The sulfate-ammonium-nitrate aerosols are also increased in the US, but OC aerosols are decreased in this region, which cancels out the effect of increased sulfate-nitrate-ammonium aerosols. The negative RF is also shown in central Africa due to increased BC and OC aerosols (Fig. 10b and c) from increased biomass burning emission in the present day (Lamarque et al. 2010) and in western tropical Africa due to increased soil dust aerosols (Fig. 10d).

Fig. 9
figure 9

Preindustrial to present-day aerosol RF calculated using GRIMs-CCM 10-year time-slice simulations for 1850s and 2000s. The mean aerosol RF is indicated in the parenthesis in upper right corner

Full size image
Fig. 10
figure 10

Preindustrial to present-day AOD change by (a) sulfate–nitrate–ammonium aerosols, (b) black carbon, (c) organic carbon, (d) soil dust aerosols, and (e) sea salt aerosols

Full size image

Figure 11 compares the global mean aerosol direct RF with the aerosol direct RF from the ACCMIP models in Shindell et al. (2013) and AEROCOM Phase II models in Myhre et al. (2013). The aerosol direct RFs from the models participating in the ACCMIP and AEROCOM Phase II are calculated in the same way as in this study (the flux change at TOA for simulation with present and preindustrial emissions). The mean aerosol direct RF from 10 ACCMIP models is − 0.26 W m-2 (stddev of 0.14). Likewise, the mean aerosol direct RF from 16 AEROCOM Phase II models is − 0.27 W m-2 (stddev of 0.15). The global mean aerosol direct RF calculated from the GRIMs-CCM simulations (− 0.30 W m-2) is close to the mean values from ACCMIP and AEROCOM Phase II. This result indicates that the GRIMs-CCM estimates the aerosol RF as a comparable level to the other climate models.

Fig. 11
figure 11

Comparison of global mean preindustrial to present-day aerosol radiative forcing from the GRIMs-CCM, ACCMIP models, and AEROCOM II models

Full size image

5 Summary and Conclusions

The SLCPs are considered the essential components for future climate projection because of the high uncertainty of their climatic impacts. The chemistry-climate model is much needed to better understand the chemistry-climate interaction. In this regard, we developed a new chemistry-climate model by coupling the GEOS-Chem’s chemistry modules into the GRIMs, GRIMs-CCM. We describe each component of the model and conduct the model evaluation to assess the new model's performance by focusing on ozone, aerosols, and aerosol radiative forcing.

We first conducted a 1-year simulation for 2005. Comparing GRIMS-CCM AOD with the satellite retrieved AOD reveals that the model captures the spatial distribution of annual-mean AOD but underestimates its magnitude by − 19%. The budget analyses for the aerosols show that removing aerosols from the atmosphere is faster in the GRIMs-CCM than in the reference model, GEOS-Chem, due to stronger cumulus convection and surface friction velocity.

The model adequately reproduces the seasonal and latitudinal variation of total column ozone. However, it exhibits a significant bias in the polar stratospheric ozone due to a cold bias of stratospheric temperature. The comparison also reveals that the model has difficulty reproducing the Antarctic ozone hole due to ignorance of complex halogen chemistry. It tells us that the future version of the GRIMs-CCM should consider the halogen species with heterogeneous chemistry to better represent the stratospheric ozone. It is further found that the model can simulate the seasonal variation of tropospheric ozone. However, the GEOS-Chem model with the assimilated meteorological fields shows generally better agreement with the observations than the GRIMs-CCM, indicating the importance of meteorological data for simulating tropospheric ozone.

The comparisons of the simulated surface PM with the in-situ measurements in East Asia, the United States, and Europe show that the GRIMs-CCM can reproduce surface PM's temporal and spatial variations with high correlation coefficients (0.79 for East Asia, 0.72 for the US, 0.52 for Europe). However, the GRIMs-CCM has difficulties in simulating soil dust aerosols showing different spatial patterns from the GEOS-Chem. The underestimation of soil dust in the Gobi Desert and drastic dust emission in northern Africa result in a low bias in East Asia and a high bias in Europe.

The aerosol radiative forcing is estimated to measure the overall climatic influence of aerosols. The preindustrial to present-day global-mean aerosol radiative forcing is − 0.30 W m-2. This value is comparable to the aerosol radiative forcing in other climate model comparison projects (− 0.26 W m-2 for ACCMIP models, − 0.27 W m-2 for AEROCOM II models).

Our results suggest that the GRIMs-CCM successfully simulates the chemical processes in the atmosphere and aerosol radiative forcing but also reveal that the model has several issues to be addressed in a future study. The update of physics schemes for the underlying atmospheric model may significantly contribute to improving chemistry simulation. Specifically, the update of the land surface model could contribute to the better simulation of soil dust aerosol, which is highly dependent on the surface soil moisture content and surface wind speed. The improvement of ozone simulation could also be fulfilled by resolving the cold temperature bias problem in the stratosphere. Shortly, the GRIMs-CCM will be further improved with the updates in the atmospheric model (Koo et al. 2022) and GEOS-Chem.