We develop a preprocessed interface to calculate physical and chemical process parameters using WRF-CMAQ output variables for simplified simulation in InMAP-China based on work from the Environmental Protection Agency (EPA) (Baker et al., 2020). Two NetCDF (Network Common Data Form) files containing the key parameters for simplified simulation are generated using the parameter interface developed here: one is at a 36 km resolution across the entire mainland of China and another is at a 4 km resolution over the Beijing–Tianjin–Hebei (BTH) region. The main step of the preprocessed interface includes meteorological and chemical variable extraction and merging, unit conversion, vertical layer mapping, physical and chemical process parameter calculation, and average processing. The hourly chemical and meteorological variable outputs from the WRF-CMAQ modeling system are converted into the annual average physical and chemical process parameters required for simplified simulation.

A NetCDF file containing the three-dimensional annually averaged parameters to characterize atmospheric advection, dispersion, mixing, chemical reaction and deposition is generated. Table 2 shows the relationship between the annually averaged parameters for simplified simulation and the original hourly variables. In InMAP-China, the annually averaged component and the deviation of wind speed to represent advection are calculated using hourly elements. The offset of wind vectors in different directions may result in some uncertainties in this process. The parameters of eddy diffusion and convective transport are pre-calculated using hourly elements, including temperature, pressure and boundary layer height, among others. The annual wet deposition rate is determined by the rainwater mixing ratio and cloud fractions. The annual dry deposition rate of particles and gaseous pollutants at the surface level is pre-calculated using friction speed, heat flux, radiation flux and land cover. The simplification of chemical reactions is different among pollutants. For NOx, NH3 and volatile organic compound (VOC) precursors, the annual average gas–particle partitioning is adopted and calculated before using the output concentrations of species from CMAQ. For SO2 pollutants, the annual oxidation rate of two major conversion pathways for SO2 is calculated using concentrations of the hydroxyl radical (HO) and hydrogen peroxide (H2O2) in CMAQ, and the conversion is estimated in InMAP-China.

To generate the meteorological and chemical parameters required by InMAP-China, a 1-year WRF-CMAQ simulation covering the entirety of mainland of China is conducted to output hourly meteorological and chemistry-related variables in 2017. Furthermore, a nested WRF-CMAQ simulation over the BTH region is also conducted and validated using observed data. The corresponded output data are used to generate the meteorological and chemical parameters required by InMAP-China for the 4 km resolution simulations in the BTH region. Tables S1 and S2 in the Supplement show the major configurations of the WRF-CMAQ modeling system. The WRF model (Skamarock et al., 2008) is driven by the National Centers for Environmental Prediction Final Analysis (NCEP-FNL) (10.5065/D6M043C6; National Centers for Environmental Prediction, 2000​​​​​​​) reanalysis data to provide the initial and boundary conditions. The meteorological fields derived from the WRF model are used to drive the CMAQ model (Appel et al., 2018) simulations. The air pollutant emissions used here include anthropogenic emissions over China derived from the Multi-resolution Emission Inventory for China (MEIC) model (http://meicmodel.org/, last access: 9 December 2021), anthropogenic emissions over the region of East Asia outside China derived from the MIX-2010 (a mosaic Asian anthropogenic emission inventory) inventory (Li et al., 2017) and biogenic emissions derived from the Model of Emissions of Gases and Aerosols from Nature (MEGAN) v2.10 model. The Carbon Bond 05 (CB05) mechanism and the aerosols 6 (AERO6) module are employed in the model simulation.

Table S3 summarizes the performance statistics of meteorological variables, including surface temperature, relative humidity and wind speed, in China in 2017, as simulated by the WRF model. The hourly observed data of major meteorological variables derived from the National Climatic Data Center (NCDC) are utilized here. The results show that the meteorological variables simulated by the WRF model agree well with the surface observations, which is consistent with previous studies (Wu et al., 2019; Zheng et al., 2015; Hong et al., 2017). The model performs well with respect to the prediction of surface temperature, with a mean bias (MB) of -0.7 K, a normalized mean bias (NMB) of -6.1 % and an R value of 0.9. The predictions of relative humidity at a height of 2 m are relatively satisfactory with an MB of 4.1 % and an NMB of 6.1 %. The predictions of wind speed at a height of 10 m are slightly overestimated, with an MB of 0.3 m s-1 and an NMB of 12.4 %, which may be caused by out-of-date United States Geological Survey (USGS) land use data employed in the model runs.

The SO2, NO2 and PM2.5 concentrations modeled across the domain agree well with the surface observations in terms of the statistical performance and monthly variations. Table S4 summarizes the performance of the statistics of major air pollutant concentrations. The nationwide annual average PM2.5 concentration simulated in China in 2017 was 42.1 µg m-3. Compared with the observed PM2.5 of 45.9 µg m-3, there are slight underpredictions, with an MB of 3.7 µg m-3 and an NMB of 8.1 %. The CMAQ model shows moderate underpredictions of the NO2 and SO2 concentrations, which may be related to the uncertainties in the emission inputs. For modeled NO2 concentrations, the MB and NMB are -4.6 µg m-3 and -13.9 %, respectively. For modeled SO2 concentrations, MB and NMB are -0.8 µg m-3 and -4.5 %, respectively. Figure S3 shows the monthly variation. The variation trend in the observed SO2, NO2 and PM2.5 concentrations can be reproduced in the CMAQ simulations.

We develop the preprocessed module to generate vector emission input for the InMAP-China simulation. This module can allocate air pollutant emissions to vertically and horizontally supply the missing parameters for the emission file and convert them into a shapefile vector format. The shapefile vector format's 36 km resolution emission data for the entirety of mainland of China and 4 km resolution data for the BTH region in 2017 are preprocessed using this module.

In this module, the emission data are preprocessed by source and altitude. The anthropogenic emissions of five sectors in China in 2017 from the MEIC inventory (http://meicmodel.org/, last access: 9 December 2021; Liu et al., 2015), the anthropogenic emissions over Asian regions outside of mainland China from the MIX-2010 inventory (Li et al., 2017) and the natural emissions estimated using the MEGANv2.10 model (Guenther et al., 2012) are employed in this study.

In more detail, the 0.3∘ gridded anthropogenic emissions for the residential, transportation and agricultural sectors are preprocessed and input to the surface layer. The gridded air pollutant emissions for the industrial sector and non-coal power plants are preprocessed for allocation to attitudes ranging from 130 to 240 m and from 130 to 890 m, respectively. The emissions from coal-fired power plants (CPPs) are preprocessed as point sources. The air pollutant emissions and the stack attribution of each unit are provided in the emission file. Because the stack attribution of the power unit is not provided in the MEIC inventory, we supplied the information in the preprocessed module based on NEI (National Emissions Inventory) data of power units (United States Environmental Protection Agency, 2011). For the stack height/stack diameter, a linear relationship is first established (see Fig. S1); supplementation of these two parameters for Chinese power plants is then conducted using the relationships. A fixed value for the other two stack attribution variables is set here because the PM2.5 concentrations attributable to power plants (CPP-PM2.5) are less sensitive to these two variables (see Fig. S2). The stack gas exit velocity and stack gas exit temperature of the power unit are 6 m s-1 and 313 K, respectively. The air pollutant emissions over Asian regions outside of mainland China and the natural emissions simulated by MEGANv2.10 are preprocessed and input to the surface layer.

To rapidly estimate the premature mortality attributable to PM2.5 exposure, we employ the exposure–response function from GEMM, developed by Burnett et al. (2018), to estimate PM2.5-related premature mortality. The PM2.5 concentration predicted by InMAP-China is also utilized in the estimation of premature mortality. Premature mortality due to noncommunicable diseases (NCDs) and lower respiratory infections (LRIs) was considered in this study. Mortality is determined by the mortality incidence rate, the population and the attributable fraction (AF) to certain PM2.5 concentrations. The national mortality incidence rate and the population data were derived from the GBD 2017 study (Institute for Health Metrics and Evaluation; Global Burden of Disease Collaborative Network, 2018a, b). The spatial distribution of the population in 2015 from the Gridded Population of World (GPW) version 4 (Doxsey et al., 2015) was employed to allocate the population in 2017.

In this study, the performance of the InMAP-China predictions are evaluated by comparison against CMAQ simulations and surface observations. Model–model and model–observation comparison have both been used to evaluate the performance of reduced-complexity air quality models in previous studies (Tessum et al., 2017; Gilmore et al., 2019).

The following aspects are considered when carrying out an evaluation. First, we examine the ability of InMAP-China to predict PM2.5 concentrations at different emission levels, which will be introduced in Sect. 3.1. We define four key regions here (see Fig. 2) and examine the performance for each region. Second, to examine the ability of the model to quantify source contributions to PM2.5 concentrations, we compare the InMAP-China predictions of the sectoral contributions attributable to power, industry, residential, transportation and agriculture with those based on the CMAQ model, which will be presented in Sect. 3.2. Third, to comprehensively understand the model's performance at a higher spatial resolution, we compare the predictions of PM2.5 concentrations at a 4 km spatial resolution in the BTH region modeled by both InMAP-China and the conventional CMAQ model with the observations (see Sect. 3.3). Fourth, focusing on the health impacts, the PM2.5-related premature mortality predicted by InMAP-China is also compared with a mortality estimation based on PM2.5 exposure derived from CMAQ, which is presented in Sect. 3.4.

For the observed PM2.5 concentration data, the annual average observed PM2.5 concentrations in 2017 were calculated using hourly concentration data from the China National Environmental Monitoring Centre (CNEMC; http://www.cnemc.cn/, last access: 9 December 2021). More than 1400 national monitoring sites for air pollutant concentrations are included in the simulation domain. The statistical parameters used in this study include the correlation coefficient (R), mean bias (MB), mean error (ME), normalized mean bias (NMB), normalized mean error (NME) and root-mean-square error (RMSE). The statistical analyses on the performance of InMAP-China are similar to our previous evaluation of conventional CTMs (Zheng et al., 2015; Wu et al., 2019).

We designed 12 simulations to examine the model ability of InMAP-China in this study. Table 3 shows the sequence of simulations.

“InMAP_TOT” represents the baseline simulation with the maximum input of combined emissions. Five sectoral anthropogenic emissions derived from the MEIC inventory, natural emissions derived from the MEGANv2.10 model and Asian emissions outside of mainland China derived from the MIX-2010 inventory are utilized in the simulation. Five sectoral and five abatement simulations are also conducted to examine the ability of InMAP-China to predict concentration changes in response to sectoral emissions and abatement emissions. The emission inputs for these 10 simulations are given in Table 3. The annual average physical and chemical process parameters are calculated based on the output variables of the WRF-CMAQ model, which has already been mentioned in Sect. 2.1.2. Based on the above input, the particle continuity equations are solved by the InMAP-China model to obtain the annual average PM2.5 concentrations at atmospheric steady state. The above simulations are all conducted at a 36km spatial resolution across the entirety of mainland of China. Furthermore, another simulation represented by “InMAP-BTH” is conducted at a 4 km spatial resolution over the BTH region using the anthropogenic emission input data at a 4 km resolution derived from the MEIC inventory and the natural emission data derived from the MEGANv2.10 model.

To carry out a comparison with the InMAP-China simulations, 11 CMAQ simulations are also performed using the same emission inputs. The hourly PM2.5 concentrations simulated by CMAQ in 2017 are averaged over the entire year. Due to limited computational resources, each simulation is conducted for four representative months (January, April, July and October) in 2017.

Figure 3 shows the performance evaluation of total PM2.5 concentrations in the InMAP_TOT simulations. Compared with the observed annual average PM2.5 concentrations, the total PM2.5 concentrations are moderately underpredicted by InMAP-China, with an MB of -8.1 µg m-3 and an NMB of -18.1 %. Compared with the CMAQ predictions, the total PM2.5 concentrations are also underpredicted, with an MB of -5.3 µg m-3, due to the underprediction of primary PM2.5. Consistent air pollutant emissions are employed in the CMAQ and InMAP-China simulations. Therefore, the underpredictions are caused by the different mechanisms in the two models. InMAP-China reproduces the spatial pattern of total PM2.5 concentrations simulated by CMAQ. Notably, significant overpredictions of PM2.5 concentrations can be observed over mountainous areas across northern China, and the complex terrain and high emission intensity increase the challenge of predicting PM2.5 concentrations using the reduced-complexity air quality model in this region.

Figure 4 shows a comparison of PM2.5 compositions. Compared with the CMAQ results, the InMAP-China predictions of PM2.5 compositions are satisfactory, with NMBs for SO42-, NO3-, NH4+ and primary PM2.5 of 13 %, -8 %, -10 % and -23 %, respectively. The predictions of SO42-, NO3- and NH4+ perform better than those of primary PM2.5. Figures 5 and 6 compare the spatial distribution of PM2.5 compositions, and similar overpredictions of PM2.5 compositions can be observed in the mountainous area in northern China.

The ability of InMAP-China to predict PM2.5 compositions is also examined at various emission levels. Figure 7 compares the concentrations of PM2.5 compositions and the proportions of secondary inorganic aerosols (hereafter, SNA) in total PM2.5 concentrations in different scenarios run by the two models. In the InMAP_TOT scenario, the proportion of SNA is 56 %, which is extremely close to the 50 % proportion in the WRF-CMAQ simulations. In five emission abatement simulations, the proportion was approximately equal to that in the baseline scenario because the linearly treated chemical reaction relationship of SNA was employed in InMAP-China. However, focusing on the five sectoral emission scenario simulations, a significant difference can be observed, which is mainly caused by the difference in the treatment of chemicals in InMAP-China and CMAQ. In this situation, the impacts on PM2.5 concentrations are distinct due to the nonlinear emission-concentration process.

Figure 8 compares the InMAP-China and CMAQ predictions of population-weighted PM2.5 concentrations and PM2.5 compositions for 11 emission scenarios. Marginal changes in air pollutant concentrations are defined as 1 µg m-3 by normalizing the population-weighted air pollutant concentrations of each scenario using the largest value among all scenarios modeled by CMAQ. InMAP-China reproduces CMAQ predictions with respect to the marginal change in population-weighted PM2.5 concentrations, with an NMB of -12 % and correlations of 0.98, as shown in Fig. 8a. This performance is similar to that predicted by InMAP in the USA (Tessum et al., 2017).

Figure 8b–f compare the predictions of PM2.5 compositions. InMAP-China predictions of SO42-, NO3-, NH4+ and primary PM2.5 agree well with the CMAQ results, but the predictions of secondary organic aerosol (SOA) are the poorest. The marginal changes in NO3- and primary PM2.5 concentrations are moderately underpredicted by InMAP-China, with NMB values of -13 % and -21 %, respectively. Conversely, the marginal change in the SO42- concentrations are overpredicted, with an NMB of 23 %. The marginal change in NH4+ predicted by InMAP-China agrees well with the CMAQ predictions. Because few reaction pathways of SOA are included in the CB05 mechanism in the CMAQ simulations, SOAs are underpredicted in the entire modeling system.

The regional performance of the changes in PM2.5 and its compositions for the 11 emission scenarios are also examined in this study. Figures S4–S7 show the regional results. Four regions, including the Beijing–Tianjin–Hebei region (BTH), the Yangtze River Delta (YRD), the Pearl River Delta (PRD) and the Fenwei Plain (FWP), are analyzed here (see Fig. 2). At the regional level, the CMAQ predicted marginal changes in population-weighted PM2.5 concentrations, and its composition can be reproduced by InMAP-China, which is similar to the nationwide performance. However, the marginal change in the SO42- concentrations over the BTH is significantly overpredicted by InMAP-China, with an NMB of 135 %, which is expected to be improved by optimizing the representation of the annual sulfate oxidation rate in this region.