Reduced-complexity air quality intervention modelling 1 over China: development of the InMAPv1.6.1-China and 2 comparison with the CMAQv5.2 model

. This paper presents the first development and evaluation of the reduced-complexity air quality 16 model for China. In this study, a reduced-complexity air quality intervention model over China ( InMAP-17 China) is developed by linking a regional air quality model, a reduced-complexity air quality model, an 18 emission inventory database for China, and a health impact assessment model to rapidly estimate the air 19 quality and health impacts of emission sources in China. The modelling system is applied over mainland 20 China for 2017 under various emission scenarios. A comprehensive model evaluation is conducted by 21 comparison against conventional CMAQ simulations and ground-based observations. We found that 22 InMAP-China satisfactorily predicted total PM 2.5 concentrations in terms of statistical performance. 23 Compared with the observed PM 2.5 concentrations, the mean bias (MB), normalized mean bias (NMB), 24 and correlations of the total PM 2.5 concentrations are -8.1 µg/m 3 , -18%, and 0.6, respectively. The 25 statistical performance is considered to be satisfactory for a reduced-complexity air quality model and 26 remains consistent with that evaluated in the United States. The underestimation of total PM 2.5 27 concentrations was mainly caused by


Introduction
With rapid urbanization and industrialization, fine particulate matter pollution less than 2.5 µm in diameter (PM 2.5 ) has become a major environmental issue in China.High PM 2.5 concentrations can be observed over eastern China from satellite observations (Xiao et al., 2020) and the PM 2.5 concentrations have been largely decreased since 2013 due to the effective control measures taken by Chinese governments ( Zhao et al., 2021).PM 2.5 can affect air quality, ecosystems, and climate change and damage human health through short-term or long-term exposure.The Global Burden of Disease study reported that 1.1 million premature deaths were caused by long-term PM 2.5 exposure over China in 2015 (Cohen et al., 2017).
State-of-the-science three-dimensional air quality models ( AQMs) have been widely used in China as tools to simulate regional PM 2.5 concentrations, quantify the contributions to total PM 2.5 concentrations resulting from emission sources and assess the benefits associated with control measures (Chang et al.;2019, Li et al., 2015;;Zhang et al., 2015;Zhang et al., 2019).The Weather Research and Forecasting model-Community Multiscale Air Quality Modelling System (WRF-CMAQ) (Appel et al., 2017;Chang et al., 2019), the Weather Research and Forecasting model coupled with Chemistry (WRF-Chem) (Reddington et al., 2019), the Weather Research and Forecasting model-Comprehensive Air Quality Model Extension (WRF-CAMx) (Li et al., 2015), and the Global Adjoint model of Atmospheric Chemistry (GEOS-Chem Adjoint) (Zhang et al., 2015) were frequently used in previous studies.To conduct a series of simulations for multiple scenarios or quantify the separate contributions attributable to multiple sources, large computational resources and run time are required while utilizing conventional AQMs.To address these challenges and to improve the availability and accessibility of air quality modelling, a number of reduced-complexity models have been developed by the air quality research community.The three representative reduced-complexity air quality models frequently used are the Estimating Air Pollution Social Impacts Using Regression (EASIUR) model (Heo et al., 2016;Heo et al., 2017), the updated Air Pollution Emission Experiments and Policy (APEEP2) model (Muller et al., 2007;Muller et al., 2011) and the Intervention for Air Pollution model (InMAP) (Tessum et al., 2017).
A recent study compares three reduced-complexity models, EASIUR, APEEP2, and InMAP, and the results indicate that these three models are consistent in their assessment of the marginal social cost at the county level (Gilmore et al., 2019).Reduced-complexity air quality models are less computationally intensive and easier to use.However, it is not available for China.Therefore, it is essential to develop a reduced-complexity air quality model over China to quickly predict PM 2.5 concentrations and the associated health impacts of emission sources.
The reduced-complexity intervention model for air pollution, InMAP, was developed by Tessum et al. (Tessum et al., 2017) to rapidly assess the air pollution, health, and economic impacts resulting from marginal changes in air pollutant emissions.Compared with conventional air quality models, InMAP has the advantage of time efficient, can predict annual-average PM 2.5 concentrations within few hours but with a modest reduction in accuracy compared with CTMs.InMAP reduces the running time by simplifying the physical and chemical process.InMAP has been used to assess marginal health damage of location-specific emission sources (Goodkind et al., 2019), to quantify the health impacts of individual coal-fired power plants in the United States (Thind et al., 2019) and to estimate the health benefits of control policies considering specific locations (Sergi et al., 2020).However, to date, a version of the reduced-complexity air quality intervention model over China is absent.
In this work, based on the source code of the version 1.6.1 of InMAP model，a reduced-complexity air quality intervention model over China ( InMAP-China) is developed to rapidly predict the air quality and estimate the health impacts of emission sources in China.The total consumed time for a simulation for the year 2017 using the InMAP-China established in this study is approximately an hour with a single CPU of 24 nodes.Therefore, it is convenient when conducting multiple simulations of PM 2.5 concentrations due to air pollutants emissions in 2017.The modelling system is applied over mainland China for 2017 under various emission scenarios to examine model performance.Comparisons against conventional air quality models and surface observations are performed in this study.The model applicability and limitations are also declared.
The paper is organized as follows: Section 2.1 presents the components of InMAP-China including the interface development between WRF-CMAQ and InMAP to generate parameters of the base atmospheric state, the preprocessed process of emission input data and the exposure-response functions employed in this model.Section 2.2 introduces the evaluation protocol, including the statistical variables adopted and the simulation design in this study.Section 3 presents the evaluation of InMAP-China's predictions of PM 2.5 air quality and PM 2.5 -related health impacts in several simulations.Section 4 summarizes the conclusions and limitations of this study.

Model components and configurations
The reduced-complexity intervention model for air pollution, InMAP, was developed by Tessum et al. (Tessum et al., 2017) to rapidly assess the air pollution, health, and economic impacts resulting from marginal changes in air pollutant emissions.The model has been widely used in studies (Sergi et al., 2020;Thind et al., 2019;Goodkind et al., 2019;Dimanchevi et al., 2019) focusing on PM 2.5 pollution and health, economic impacts resulting from emission sources in the United States.In this model, the continuous equation of atmospheric pollutants is solved at an annual scale, and the run time can be reduced.The parameters used to represent physical and chemical processes for simplified simulation are calculated prior to using CTM output data.PM 2.5 air quality and PM 2.5 -related premature mortality are predicted and output in the InMAP model.

In this work, a Chinese version of the reduced-complexity air quality intervention model InMAP-
China is developed for the purpose of rapidly estimating the PM 2.5 concentration and associated health impacts of emission sources.Figure 1 shows the model framework.Based on the source code of the InMAP model, three-step development work is conducted to establish InMAP-China.First, we develop a preprocessed interface to calculate physical and chemical process parameters using the WRF-CMAQ output variables to support the simplified simulation in InMAP-China.Second, air pollutant emission data are preprocessed to an appropriate format for the InMAP-China simulation.Third, the exposureresponse function of the GEMM model is employed in InMAP-China and replaces the original default function to assess PM 2.5 -related health impacts.
Table 1 presents the basic configurations of InMAP-China.The simulation domain is over East Asia and covers mainland China.The spatial resolution is 36 km.Fourteen vertical layers are used in InMAP-China, ranging from the surface layer to the top level of the tropospheric layer.

Parameter interface development for simplified simulation in InMAP-China
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 the Environmental Protection Agency's (EPA) work (Baker et al., 2020).Two NETCDF files containing the key parameters for simplified simulation are generated by using the parameter interface developed here, one is at 36km resolution across entire mainland of China and another is at 4km resolution over the 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 modelling system are converted into 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 annual-average parameters for simplified simulation and the original hourly variables.In InMAP-China, the annual 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 precalculated using hourly elements, including temperature, pressure, boundary layer height, etc.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 precalculated using friction speed, heat flux, radiation flux and land cover.The simplification of chemical reactions is different among pollutants.For NO x , NH 3 , and volatile organic compound (VOC) precursors, the annual averaged gas-particle partitioning is adopted and calculated before using the output concentrations of species from CMAQ.For SO 2 pollutants, the annual oxidation rate of two major conversion pathways for SO 2 is calculated using concentrations of hydroxyl radical (HO) and hydrogen peroxide (H 2 O 2 ) in CMAQ, and the conversion is estimated in InMAP-China.

Prior WRF-CMAQ simulation
To generate the meteorological and chemical parameters required by InMAP-China, a one-year WRF-CMAQ simulation covering the entire mainland of China is conducted to output hourly meteorological and chemical-related variables in the year 2017.Besides, the nested WRF-CMAQ simulation over the BTH region is also conducted and validated using observed data.The corresponded output data is used to generate the meteorological and chemical parameters required by InMAP-China for the simulations of 4 km resolution in the BTH region.Tables S1 and S2 show the major configurations of the WRF-CMAQ modelling system.The WRF model is driven by the National Centers for Environmental Prediction Final Analysis (NCEP-FNL) (https://doi.org/10.5065/D6M043C6)reanalysis data to provide the initial and boundary conditions.The meteorological fields derived from the WRF model is used to drive the CMAQ model (Appel et al., 2016) simulations.The air pollutant emissions used here include anthropogenic emissions over China derived from the MEIC model (http://meicmodel.org/),anthropogenic emissions over the region of East Asia outside China derived from the MIX-2010 inventory (Li et al., 2015), and biogenic emissions derived from the MEGANv2.10 model.The CB05 chemical mechanism and the AERO6 aerosol 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 Climate 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 on the predictions of surface temperature, with an MB of -0.7 K, an NMB of -6.1%, and R of 0.9.The predictions of relative humidity at a height of 2 metres are relatively satisfied with an MB of 4.1% and an NMB of 6.1%.The predictions of wind speed at a height of 10 metres are slightly overestimated, with an MB of 0.3 m/s and an NMB of 12.4%, which may be caused by out-of-date USGS land use data employed in the model runs.
The SO 2 , NO 2 and PM 2.5 concentrations modelled 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 averaged PM 2.5 concentration simulated in 2017 in China was 42.1 µg/m 3 .Compared with the observed PM 2.5 of 45.9 µg/m 3 , there are slight underpredictions with an MB of 3.7 µg/m 3 and NMB of 8.1% .The CMAQ model has moderate underpredictions of the NO 2 concentrations and SO 2 concentrations, which may be related to the uncertainties of emission inputs.For modelled NO 2 concentrations, MB and NMB are -4.6 µg/m 3 and -13.9%, respectively.For modelled SO 2 concentrations, MB and NMB are -0.8 µg/m 3 and -4.5%, respectively.Figure S3 shows the monthly variation.The variation trend of the observed SO 2 , NO 2 , and PM 2.5 concentrations can basically be reproduced in the CMAQ simulations.

Preprocessed emission input data
We develop the preprocessed module to generate vector emission input for the InMAP-China simulation.This module can allocate air pollutant emissions vertically and horizontally to supply the missing parameters for the emission file and convert them into a shapefile vector format.The shapefile vector format's emission data of 36km resolution in entire mainland of China and 4km resolution in the BTH region in 2017 are pre-processed by 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/), the anthropogenic emissions over regions outside mainland China in Asia from the MIX-2010 inventory (Li et al., 2015), and the natural emissions estimated using the MEGANv2.10 model (Guenther et al., 2012) are employed in this study.
More detailed, the gridded anthropogenic emissions of 0.3 degrees for the residential, transportation, and agricultural sectors are preprocessed and input to the surface layer.The gridded air pollutant emissions of the industrial sector and noncoal power plants are preprocessed for allocation to attitudes ranging from 130 metres to 240 metres and 130 metres to 890 metres, respectively.The emissions of 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 missed in the MEIC inventory, we supplied the information in the preprocessed module based on NEI ( National Emission Inventory data) data of power units.For stack height/stack diameter, a linear relationship is first established (see Figure S1), and then, supplementation for these two parameters of Chinese power plants is conducted by using the relationships.The fixed value for the other two variables of stack attribution is set here because the PM 2.5 concentrations attributable to power plants (CPPs-PM 2.5 ) are less sensitive to the two variables (see Figure S2).The stack gas exit velocity and stack gas exit temperature of the power unit are 6 m/s and 313 K, respectively.The air pollutant emissions over regions outside mainland China in Asia and the natural emissions simulated by MEGANv2.10 are preprocessed and input to the surface layer.

Exposure-response function from GEMM
To rapidly estimate the premature mortality of PM 2.5 exposures, we employ the exposure-response function from GEMM to estimate PM 2.5 -related premature mortality, which was developed by Burnett et al. (Burnett et al., 2018), and calculate the premature mortality using PM 2.5 concentration predictions of InMAP-China.Premature mortality due to non-communicable diseases (NCDs) and lower respiratory infections (LRIs) was considered in this study.Mortality is determined by the mortality incidence rate, population, and attributable fraction (AF) to certain PM2. 5 concentrations.The national mortality incidence rate and the population data were derived from the GBD2017 study (Institute for Health Metrics and Evaluation).The spatial distribution of the population in 2015 from the Gridded Population of World Version 4 (Doxsey et al., 2015) was employed to allocate the population in 2017.

Evaluation method
In this study, the performances of the InMAP-China predictions are evaluated by comparison against CMAQ simulations and surface observations.Model-to-model comparison and model-toobservation 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 to make an evaluation.First, we examine the ability of InMAP-China to predict PM 2.5 concentrations at different emission levels, which will be introduced in Section 3.1.Second, to examine the ability to quantify source contributions to PM 2.5 concentrations, we compare the InMAP-China's 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 Section 3.2.Third, to comprehensively understand the performance at higher spatial resolution using InMAP-China, we compare the predictions of PM 2.5 concentrations at 4km spatial resolution in the BTH region both modelled by InMAP-China and conventional CMAQ with the observations, which is displayed in Section 3.3.Fourth, focusing on the health impacts, the PM 2.5 -related premature mortality predicted by InMAP-China is also compared with mortality estimation based on PM 2.5 exposure derived from CMAQ, which is presented in Section 3.4.
For the observed PM 2.5 concentration data, the annual averaged observed PM 2.5 concentrations in 2017 were calculated using hourly concentration data from the China National Environmental Monitoring Center, CNEMC (http://www.cnemc.cn/).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).

Experimental design
We design twelve 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 maximum emissions input, in which five sectoral anthropogenic emissions are derived from the MEIC inventory, natural emissions are derived from the MEGANv2.10 model, and Asian emissions outside mainland China are derived from the MIX-2010 inventory are combined as emission inputs.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 ten simulations have been declared in Table 3.The annual averaged physical and chemical process parameters are calculated based on the output variables of WRF-CMAQ model, which has already been mentioned in Section 2.1.2.
Based on the above input, the particle continuity equations are solved by InMAP-China model to obtain the annual averaged PM 2.5 concentrations at the steady-state of the atmosphere.The above simulations are all conducted at 36km spatial resolution across the entire mainland of China.Besides, another simulation represented by InMAP-BTH is conducted at 4km spatial resolution over the BTH region, with the anthropogenic emission input data at 4km resolution derived from the MEIC inventory and natural emissions derived from the MEGANv2.10 model is utilized in this simulation.
In order to make a comparison with the InMAP-China simulations, eleven CMAQ simulations are also performed under the same emission inputs.The hourly PM 2.5 concentrations simulated by CMAQ in 2017 are averaged at obtain the annual averaged PM 2.5 concentrations.Due to limited computational resources, each simulation is conducted for four representative months (January, April, July, and October) in 2017.

Total PM 2.5 concentrations
Figure 3 shows the performance evaluation of total PM 2.5 concentrations in the InMAP_TOT simulations.Compared with the observed annual averaged PM 2.5 concentrations, the total PM 2.5 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 PM 2.5 concentrations are also underpredicted, with an MB of -5.3 µg/m 3 due to the underprediction of primary PM 2.5 .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.Basically, InMAP-China reproduces the spatial pattern of total PM 2.5 concentrations simulated by CMAQ.Notably, significant overpredictions of PM 2.5 concentrations can be observed over mountain areas across Northern China, and the complex terrain and large emission intensity increase the challenge of predicting PM 2.5 concentrations using the reducedcomplexity air quality model in this region.
Figure 4 shows a comparison of PM 2.5 compositions.Compared with the CMAQ results, the InMAP-China predictions of PM 2.5 compositions are satisfactory, with NMBs for SO 4 2-, NO 3 -, NH 4 + , and primary PM 2.5 equal to 13%, -8%, -10%, and -23%, respectively.The predictions of SO 4 2-, NO 3 -, and NH 4 + perform better than those of primary PM 2.5 .Figure 5 and Figure 6 compare the spatial distribution of PM 2.5 compositions, and similar over-predictions of PM 2.5 compositions can be observed in the mountain area in Northern China.
The ability of InMAP-China to predict PM 2.5 compositions is also examined at various emission levels.Figure 7 compares the concentrations of PM 2.5 compositions and the proportions of secondary inorganic aerosols (hereafter, SNA) in total PM 2.5 concentrations in different scenarios by 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 simulations of five sectoral emission scenarios, a significant difference can be observed, which is mainly caused by the difference in chemical treatments in InMAP-China and CMAQ.In this situation, the impacts on PM 2.5 concentrations are distinct due to the nonlinear emission-concentration process.concentrations, with a NMB of -12% and correlations of 0.98, as shown in Figure 8(a).This performance is similar to that predicted by InMAP in the United States (Tessum et al., 2017).are underpredicted in the entire modelling system.
The regional performance of the changes in PM 2.5 and its compositions for eleven emission scenarios is also examined in this study.Figures S4-S7 show the regional results.Four regions, including the Beijing-Tianjin-Hebei region ( BTH), Yangtze River Delta ( YRD), Pearl River Delta ( PRD), and Fen Wei Plain ( FWP), are analysed here (see Figure 2).At the regional level, the CMAQ predicted marginal changes in population-weighted PM 2.5 concentrations, and its composition can be reproduced by InMAP-China, which is similar to the nationwide performance.However, the marginal change in SO 4 2-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.

Model performance of source contributions in China
Figure 9 shows the contribution of each sector to PM 2.5 concentrations nationwide and at the regional scale, and Table 4 displays the proportion value of sectoral contribution based on two models.The predictions of the source contributions of PM 2.5 concentrations in InMAP-China are basically reliable compared with those based on the CMAQ model, and the difference can be explained.
The results based on the two models indicate that the industrial and residential sectors are the first and second contributors among the five sectors.The contribution of the electricity sector is comparable when using the two models, while the contributions of transportation and agriculture are moderately different, which is mainly due to the difference in the model mechanism and the treatment of secondary inorganic aerosols in the two models.At the regional scale, the difference in the sectoral contribution caused by the mechanism in the two models is more significant than at the national scale.

Model performance of PM 2.5 predictions at higher resolution in the BTH region
We also conducted a simulation with higher spatial resolution of 4 km in the BTH region by using InMAP-China model and make a comparison with the WRF-CMAQ nested simulation at the same area in the BTH region.Figure 10 and Figure 11 show the performance evaluation of total PM 2.5 concentration and the composition in the InMAP_BTH scenario.Compared with the observed annual averaged PM 2.5 concentrations, the total PM 2.5 concentrations are moderately overpredicted in InMAP_BTH with an NMB of 41.3% and an R of 0.5.
Further compared with the nested CMAQ predictions, the total PM 2.5 concentrations are also overpredicted by InMAP-China model.The predictions of PM 2.5 compositions in the InMAP_BTH scenario are partially satisfactory, except for SO 4 2-, with NMBs for SO 4 2-, NO 3 -, NH 4 + , and primary PM 2.5 equal to 178%, 36%, 33%, and 27%, respectively.Figure 12 further shows the comparison of the spatial distribution of PM 2.5 compositions in the BTH region.The overall spatial distribution pattern of PM 2.5 compositions is similarly modeled by two models, however, an obvious difference can be observed across the mountain area in the BTH region, for instance, the over-predictions of PM 2.5 compositions, especially, SO 4 2-and NO 3 -observed near the Taihang mountain area.

Model performance of PM 2.5 -related premature mortality in China
To examine the performance of the predictions of PM 2.5 -related premature mortality, a comparison of premature mortality using the PM 2.5 predictions from InMAP-China and CMAQ, separately, is performed here.Figure 13 shows the comparison based on two models for all provinces.The results demonstrate that, compared with the premature mortality based on CMAQ, the relative difference is ranging from -44% to 15% at the provincial level due to the difference of PM 2.5 concentrations in the two models.
At the provincial level, the PM 2.5 -related premature mortality in Beijing city, Tianjin city, Hebei province, and Shanghai city is slightly over-predicted by InMAP-China, with the relative difference ranging from 4% to 15%.Conversely, for the other majority of provinces, PM 2.5 -related premature mortality is under-predicted by InMAP-China, with the relative difference ranging from -3% to -44%.
Overall, the PM 2.5 -related premature mortality estimated using InMAP-China was 1.92 million people in 2017.Compared with the CMAQ-based estimations, 25 ten thousand deaths are under-predicted by InMAP-China because of underestimation of total PM 2.5 concentrations in the baseline simulation.

Conclusions
This work develops a reduced-complexity air quality intervention model over China and presents a comprehensive evaluation by comparing CMAQ simulations and surface observations.The InMAP-China aims at providing a simplified modeling tool to rapidly predict the PM 2.5 concentrations due to emission change as well as health impact of emission sources in China.After the model is established, the total consumed time for a new simulation under the atmosphere condition in the year 2017 across the mainland of China using InMAP-China is merely an hour with a single CPU of 24 nodes.Therefore, it is time-efficient when conduct new simulations of PM 2.5 concentrations in China.Notably, the running of WRF-CMAQ simulations is merely necessary in our developing stage of InMAP-China.For the application of InMAP-China, we recommend users to select InMAP-China as a prior tool with extensive simulation demands, for instance, to quantify the PM 2.5 concentrations due to hundreds of pollution emitters or to rapidly estimate the PM 2.5 concentrations caused by dozens of control policies, separately.
Besides, the variable grid can also be set in InMAP-China to allow high spatial resolution of 1km or even higher in certain urban area.
InMAP-China has moderately satisfactory performance in this study, however, this model has reductions in accuracy compared with conventional CTMs.Overall, InMAP-China satisfactorily predicts total PM 2.5 concentrations in the baseline simulation in terms of statistical performance.Compared with the observed PM 2.5 concentrations, the MB, NMB, and correlations of the total PM 2.5 concentrations are -8.1 µg/m 3 , -18%, and 0.6, respectively.The statistical performance is satisfactory for a reducedcomplexity air quality model and remains consistent with the performance evaluation in the United States.
The underestimation of total PM 2.5 mainly comes from the primary PM 2.5 .Moreover, the spatial pattern of total PM 2.5 concentrations can be reproduced in InMAP-China, while an overestimation over the mountain area in Northern China can be observed.The large emission intensity and complex terrain over this region increase the difficulty of modelling concentrations in this area.The predictions of source contributions to PM 2.5 concentrations by InMAP-China are comparable with those based on the CMAQ model, and the difference is mainly caused by the uncertainty of the simplification of chemical process in the InMAP-China.The global version of reduced-complexity air quality model ( Global-InMAP) is also developed and preprint recently ( Thakrar et al., 2021), our results of InMAP-China can provide more accurate result in the mainland of China.
This study is subject to some limitations and uncertainties.In InMAP-China, the annual-average chemical and physical processes parameters are calculated using hourly parameters from WRF-CMAQ.
Complicated seasonal and daily variations affecting the formation and transportation of particulate matter are challenging to retain.The intensity of advection of the air mass is supposed to be weakened due to the offset of the wind vector in the averaging process, which was also pointed out in a previous study.

Figure 8
Figure 8 compares the InMAP-China and CMAQ predictions of population-weighted PM 2.5 concentrations and PM 2.5 compositions for eleven 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 modelled by CMAQ.The InMAP-China reproduces CMAQ predictions on the marginal change in population-weighted PM 2.5

Figure 8
Figure 8(b)-(f) compares the predictions of PM 2.5 compositions.The InMAP-China predictions of SO 4 2-, NO 3 -, NH 4 + and primary PM 2.5 agree well with the CMAQ results, but the predictions of secondary organic aerosol (SOA) are the poorest.The marginal changes in NO 3 -and primary PM 2.5 concentrations are moderately underpredicted by InMAP-China, with NMB values of -13% and -21%, respectively.Conversely, the marginal change in SO 4 2-concentrations is overpredicted with an NMB of 23%.The marginal change in NH 4 + 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

Figure 2 29 Figure 3 Figure 4
Figure 2 Four key regions defined in this study, including the Beijing-Tianjin-Hebei region, Yangtze River Delta region, Pearl River Delta region and Fen Wei Plain region.

Figure 5
Figure 5 The spatial pattern of PM 2.5 compositions modelled by the InMAP-China and WRF-CMAQ models.

Figure 6
Figure 6 The difference in the spatial pattern of PM 2.5 compositions between InMAP-China and WRF-CMAQ.

Figure 7
Figure 7 Comparison of PM 2.5 component concentrations and SNA contributions in these eleven simulations.

Figure 9
Figure 9 Comparison of source contributions to population-weighted PM 2.5 concentrations estimated by the

Figure 10
Figure 10 Scatter plot comparing the PM 2.5 concentration modeled in the BTH region with 4 km spatial

Figure 11
Figure 11 Scatter plot comparing the PM 2.5 composition concentration modeled at BTH region with 4km

Figure 12
Figure 12 The spatial pattern of PM 2.5 compositions simulated in the BTH region with 4km spatial resolution

Figure 13
Figure13Comparison of PM 2.5 -related premature mortality using the PM 2.5 predictions from two models.

Table 1 . Model configurations in InMAP-China.
Moreover, InMAP-China has difficulty predicting SOA concentrations because reaction pathways for SOA are insufficient in this modelling system.Further research work is suggested to improve the model performance.For instance, the combination of machine learning with the simplified simulation may need to research to promote the reduced-complexity air quality modeling over China.formation during the January 2013 haze episode in North China.Atmospheric Chemistry and Physics,