Sensitivity Analysis of Chemical Mechanisms in the WRF-Chem Model in Reconstructing Aerosol Concentrations and Optical Properties in the Tibetan Plateau

To investigate the effect of gas-phase chemical schemes and aerosol mechanisms on the reconstruction of the concentrations and optical properties of aerosols in the Tibetan Plateau (TP) and adjacent regions, two simulation experiments using the mesoscale Weather Research and Forecasting (WRF) meteorological model with the chemistry module (WRF-Chem) were performed in 2013. The RADM2 gas-phase chemical mechanism and the MADE/SORGAM aerosol scheme were selected in the first configuration, whereas the CBMZ gas and MOSAIC aerosol reaction schemes were included in the second simulation. The comparison demonstrated that chemical mechanisms play a key role in affecting the evolution of gas-phase precursors and aerosol processes. Specifically, compared with RADM2, CBMZ revealed lower O3 and higher NO2 surface concentrations, because of more efficient O3-NO titration, and higher HNO3 concentrations owing to more effective NO2 + OH reaction. SO2 could easily form particulate sulfate through cloud oxidation in RADM2. The MADE/SORGAM module presented higher surface PM2.5 and PM10 concentrations than did the MOSAIC module over the TP and in surrounding regions, because of the difference in aerosol compounds and the distribution of computed aerosol concentrations between modes and bins. The aerosol optical depth at 550 nm indicated a potential correlation with surface secondary inorganic aerosols concentrations. Higher surface sulfate and nitrate concentrations appeared to determine higher AOD values in MADE/SORGAM than in MOSAIC. Finally, the comparison with observations suggested that, the simulation performed using the CBMZ gas-phase chemical mechanism and MOSAIC aerosol module could be suitable for the efficient reconstruction of aerosols and their optical depth over the TP.


INTRODUCTION
Aerosol is a colloid of solid or liquid particles in a gas, mainly containing sulfate, ammonium, nitrate, organic carbon, black carbon, sea salt, and dust.Aerosols can affect radiative transfer directly by absorbing and scattering solar radiation and indirectly by modifying the microphysical and optical properties of clouds (Kiehl et al., 1993;Jones et al., 1994), thereby resulting in a cooling (e.g., sulphate (Andreae et al., 2005)) or heating (e.g., black carbon (Jacobson, 2001)) effect on the planet.In contrast to greenhouse gases, aerosols are not well-mixed, therefore, they are vulnerable to be affected by meteorological conditions, chemical mechanisms, and aerosol modules, leading to substantial heterogeneity in their spatial and temporal variations, optical properties, and subsequent radiative effect (Huang et al., 2009, Ji et al. 2016).
The Tibetan Plateau (TP), also referred to as the third pole because it contains the world's third-largest store of ice, has a considerable effect on radiative budgets and climate (Ye and Wu, 1998;Ma et al., 2009).According to instrumental temperature records and re-analysis datasets, the temperature in the TP significantly increased over the last 60 years (Kang et al., 2010).Therefore, glaciers have retreated rapidly in many parts of the TP (Yao et al., 2012).
Gas emissions, considered as the main cause of global warming, increasing the concentrations of aerosols (e.g., black carbon) in ambient regions appear to be another crucial anthropogenic driving force of these changes over the TP (Ramanathan and Carmichael, 2008;Lau et al., 2010;Ji et al., 2011;Cong et al., 2015;Ji et al., 2015;Kang et al., 2016;Ming et al., 2016).
The TP is located at the convergence of the two largest anthropogenic aerosol emission sources: southern Asia and eastern Asia.Once aerosols are transported to the TP, they may affect the thermal structure of the atmosphere, radiative balance, and surface albedo and further change hydrological cycle (Ji, 2016) and monsoon cycles in Asia (Huang et al., 2009;Xu et al., 2009;Lau et al., 2010;Ji et al., 2011).However, due to the harsh environment, difficult field works access, and limited instrumental stations, the measurement records are insufficient to quantitative study the spatial and temporal variations of aerosols and the climatic effects of aerosols over the TP and surrounding regions.
The mesoscale Weather Research and Forecasting (WRF, Grell et al., 2005) meteorological model with the Chemistry module (WRF-Chem) is a new generation of online-coupled meteorology and chemistry models.This model significantly reduces the inconsistency between meteorological and chemical processes and also takes feedback effects related to aerosol into account (Zhang et al., 2010;Chen et al., 2017).Nevertheless, the accurate modeling of aerosol feedback effects requires a reliable reproduction of the main driving process of the fate of atmospheric aerosols.A recent simulation study in air quality still showed considerable uncertainties in aerosol treatments (Solazzo et al., 2012), inducing significant discrepancies in modeling of aerosol concentrations and optical properties.Chemical mechanisms play a key role in the reconstruction of aerosols, which affects aerosol processes and the evolution of gas-phase precursors.
The WRF-Chem model enables the comparison and evaluation of different gas-phase chemistry and aerosol modules under the same meteorological driving data and chemical emissions (Balzarini et al., 2014).Balzarini et al. (2015) reported out that CMBZ showed higher gas concentration than did RADM2, and that MADE-SORGAM reproduced higher aerosol values than did MOSAIC over Europe in 2010.Yun et al. (2014) evaluated the sensitivity of aerosol schemes and dust options to predict particulate matters, and the scenario under the RADM2 gas-phase scheme and MADE/SORGAM aerosol module presented higher concentration values than did other scenarios.These studies mainly focused on low-altitude regions.However, the sensitivity of the WRF-Chem model to chemical mechanisms in high-altitude areas, such as the TP remains unclear.
In this study, two 1-year simulation experiments of the WRF-Chem model were designed for the TP and surrounding regions.We not only investigated the ability of different chemical mechanisms in reconstructing the concentrations and optical properties of aerosol, but also analyzed the reasons for the inconsistent performance levels of various schemes.

WRF-Chem Model and Experimental Setup
The WRF-Chem model simulates the emission, transport, mixing, and chemical transformation of trace gases and aerosols simultaneously with the meteorology.In this study, two experiments with different combinations of chemical mechanisms and aerosol modules were designed to investigate the modeling sensitivity levels of the WRF-Chem model version 3.6 over the TP.The first experiment, RS, selected the RADM2 (Regional Acid Deposition Model v2; Stockwell et al., 1990) gas-phase chemical mechanism and the MADE/SORGAM (Modal Aerosol Dynamics Model for Europe/Secondary Organic Aerosol Model; Ackermann et al., 1998;Schell et al., 2001) aerosol module, whereas the second experiment CB selected the CBMZ (Carbon Bond Mechanism version Z; Zaveri and Peters, 1999) gasphase chemistry and the MOSAIC (Model for Simulating Aerosol Interactions and Chemistry; Zaveri et al., 2008) aerosol reaction scheme.Additionally, to analyse the impact of aerosol feedbacks on the simulation differences for the meteorological variables, we set up another experiment that has no aerosol feedbacks.
The RADM2 gas-phase mechanism, a condensed gasphase photo-oxidation mechanism developed by Stockwell et al. (1990), uses a "lumped molecule" technique in which similar organic compounds are grouped together in different model categories (Grell et al., 2005), and it comprises 157 gas-phase reactions and 63 chemical species.The MADE/ SORGAM aerosol module divides aerosols into three lognormally modal distributions: the Aitken mode (< 0.1 µm diameter), the accumulation mode (0.1-2 µm diameter), and the coarse mode (> 2 µm diameter).Under this mechanism, aerosols are assumed to be internally mixed in the same mode but externally mixed among different modes (Zhao et al., 2010).
The two WRF-Chem model configurations shared the same physical parameterizations, namely the Morrison 2moment microphysical parameterization (Gustafson et al., 2007), RRTMG longwave and shortwave radiation scheme (Zhao et al., 2011), Mellor-Yamada-Janjic (MYJ) planetary boundary layer scheme (Schaefer, 1990), Noah land surface (Chen and Dudhia, 2001), and Grell-Devenyi cumulus scheme (Grell et al., 2002).WRF-Chem simulations were performed in 2013 mainly covering the TP and south Asia (Fig. 1).The first 6 days were used for model spin-up.The spatial configuration employed consists of one-way domain centered on latitude 33°N, and longitude 88°E, with horizontal resolution of 15 km, making 216 and 148 grid cells in west-east and north-south directions.There are 30 vertical sigma levels with the top at 50 hPa.The Lambert projection has been used according to the project specifications.The simulation was driven by the 6-h National Centers for Environmental Prediction (NCEP) final reanalysis data with 1° × 1° spatial resolution.
In the WRF-Chem model, improvement in modeling meteorological fields can lead to better performance in atmospheric chemistry simulation.Therefore, we added the Jimenez subgrid-scale orography parameterization scheme (Jiménez and Dudhia, 2012) and used the 2013 MODISbased land-use data to replace the default 2001 land-use data (Fig. S1, from 2001 to 2013, land-use types are shown as marked changes over the TP).The 2013 MODIS-based landuse data could improve WRF modeling for meteorological fields over the northeastern TP (Yang et al., 2016).Additionally, we assimilated AMSUA /B radiance data (NOAA-15/16/17/18/19) by the WRF-3DVar assimilation system to improve the model's initial field, according to our previous study (Yang et al., 2016).

Emissions
Anthropogenic emissions of CO, VOC (Volatile Organic Compounds), NO x , BC (black carbon), OC (organic carbon), SO 2 , PM 2.5 and PM 10 are based on Intercontinental Chemical Transport Experiment-Phase B (INTEX-B) with spatial resolution of 0.5 × 0.5 degree.As shown in supplement Fig. S2, the emission fluxes of the main anthropogenic aerosols components has low concentrations in the Tibetan Plateau but high values in South Asia and east China.The greenhouse gases emission database (e.g., CO 2 , CH 4 , and N 2 O) were derived from the Reanalysis of the TROpospheric chemical composition (RETRO, http://retro.enes.org/index.shtml) with 0.5 × 0.5 degree resolution.The sea salt and volcanic ash emissions to the study domain is interpolated from a global emissions dataset (Freitas et al. 2011).The fire emissions inventory was based on the fire inventory from NCAR (FINN) (Wiedinmyer et al., 2010).Biogenic emission is from the Model of Emission of Gases and Aerosol from Nature (MEGAN, Guenther et al., 2006).Additionally, the mozbc utility and the Model for OZone and Related chemical Tracers (MOZART, http://www.acom.ucar.edu/wrf-chem/mozart.shtml, Emmons et al., 2010) data were used to create improved initial and lateral chemical boundary conditions.

Observation Data
In the study area, 231 national observation stations from the China Meteorological Data Network (CMDN, http://data.cma.cn/) were included to evaluate the model simulation of 2-m temperature (T2), 2-m relative humidity (RH2), and 10-m wind speed (U10) in 2013.Fig. 1 showed the 231 national stations with black dots.Additionally, for the spatial pattern, dataset from the Climate Research Unit (CRU, Mitchell and Jones, 2005) was used to validate the simulated surface air temperature, and modeled surface relative humidity.Simulated 500 hPa wind was evaluated by the ERA-Interim data provided by the European Centre for Medium-Range Weather Forecasts (Dee et al., 2011).
The near surface concentration of gaseous species from six in situ observation stations and aerosol compounds from seven in situ observation stations were collected for testing the model output.The identification of Aerosol Optical Depth (AOD) over the region is based on quality-assured data from the Aerosol Robotic Network (AERONET), which is a network of sum-and sky-scanning ground-based automated radiometers providing reliable and continuous data on optical properties of various types of aerosols worldwide (Holben et al., 2001;Dubovik et al., 2002).In this study, we used the level 2.0 AOD data from 2013.Fig. 1 presented the fifteen AERONET sites with red dots.Specifically, at Namco site, the AOD data from AERONET were provided by Cong et al. (2009).

Model Performance in Meteorological Fields
As shown in Figs.2(a)-2(c), the T2 modeled by two WRF-Chem experiments modeled was in proximity to the T2 reported in the CRU dataset in 2013.The T2 was below 0°C over the TP.Warmer areas surrounded the TP, such as South Asia, which had a T2 of above 24°C.In contrast to the T2, the RH2 over the TP was higher than that in adjacent regions (Figs.2(d)-2(f)).The model showed favorable spatial consistency with the ERA-Interim reanalysis data in terms of the RH2.Compared with the CB experiment (Fig. 2(s)), the RS experiment showed higher RH2 in high-value centers, namely the southeastern and northwestern TP (Fig. 2(d)).As presented in Figs.2(g)-2(i), westerly winds prevailed over the study area and partially divided into two branches due to the topography.One branch flowed westward, and the other branch was forced by a high terrain and followed a northwesterly path.The two branches converged to form westerly winds at longitude of approximately 95°E.In South Asia, the simulated U10 in CB (Fig. 2(g)) was higher than that in RS (Fig. 2(h)).
Additonally, we analyzed the aerosol-induced mean changes in temperature and relative humidity at near-surface (Fig. S3) and vertical profile (Fig. S4).The differences of aerosol-induced mean changes in 2-m temperature between the CB and RS were shown in Fig. S5.The CB with lower 2-m temperature increment (Fig. 5 led to increased temperature in lower atmosphere but reduced temperature in upper atmosphere.There is an inverse correlation between temperature and humidity in both Figs.S3 and S4, because the decrease (increase) in temperature can lead to a decrease (increase) in the saturation pressure of water vapor thus an increase (decrease) in relative humidity.
To quantitatively evaluate model performance in surface meteorological elements, the meteorological data of 231 national stations were used.Fig. 3 presents a box plot of the daily mean variation of observed and simulated T2, RH2, and U10 at meteorological stations.Compared with the observation, the two simulation experiments showed a slightly higher T2 from January to March and a lower T2 from July to September (Fig. 3(a)).As shown in Fig. 3(b), the model can represent the annual variance of the RH2 and showed underestimation from January to March and overestimation from July to September.Considering the subgrid-scale orography, the model performance in simulating the U10 was not as satisfactory as that in simulating the T2 and RH2.The CB experiment displayed higher U10 than did the RS experiment (Fig. 3(c)).The corresponding statistics between observations and simulations are shown in Table 1.

Sensitivity Study and simulation Evaluation in Gaseous Species, Aerosol Compounds, and Optical Properties Gaseous Species
The spatial pattern of the annual mean concentrations of gas-phase compounds leading to secondary inorganic aerosols (SIA) is presented in Fig. 4. The differences between CB and RS are calculated by subtracting the substance concentrations of RS from that of CB.The CB and RS experiments revealed a quite consistent performance in the spatial distribution of gas-phase species.Surface O 3 concentrations in the CBMZ gas-phase scheme were generally lower (up to 5 ppb) than those in the RADM2 scheme (Fig. 4(a)).The difference in the surface NO 2 concentration is mostly associated with the discrepancy in the O 3 concentration (Fig. 4(b)).The O 3 and NO 2 concentrations are strongly affected by the O 3 -NO titration process (Pirovano et al., 2012).The CBMZ scheme appeared to present more efficient O 3 -NO titration than did the RADM2 scheme; thus, the CB experiment showed a higher NO 2 concentration than did the RS experiment, particularly in South Asia (up to 1.2 ppb) with high concentrations of NO x (Fig. S2).
The annual mean SO 2 concentrations in the CB experiment was generally higher than that in the RS experiment (Fig. 4(c)).This discrepancy may be related to the chemical oxidation processes of SO 2 , because the two experiments shared the same anthropogenic emissions.SO 2 was more likely to form particulate sulfate through cloud oxidation in the RADM2 gas-phase scheme (Brugh et al., 2010).Therefore, a decrease in the surface SO 2 concentration in the RS experiment can be expected.Furthermore, we analyzed the distribution of H 2 O 2 , which is one of the most efficient oxidants of sulfuric compounds in fogs and clouds (Seinfeld and Pandis, 1998).As shown in Fig. 4(d), the CB experiment showed a higher concentration of H 2 O 2 (up to 0.4 ppb) with respect to regions having a lower SO 2 concentration, suggesting inefficient SO 2 aqueous oxidation.Additionally, we compared the vertical distribution of SO 2 between the CB and RS experiments.As shown in Fig. S6(c), the mean SO 2 concentrations in the CB experiment was generally higher than that in the RS experiment.
The CB experiment generally predicted higher HNO 3 concentration than did the RS experiment, and the largest differences appeared in south Asia (Fig. 4(e)).This can be due to different reaction coefficients in the photochemical reaction of NO 2 with OH.When the temperature reached 300 K, the constant of reaction rate in NO 2 + OH + M → HNO 3 was approximately 1.3 times higher in the CBMZ scheme than in the RADM2 scheme (Balzarini et al., 2015).Furthermore, in vertical direction, the HNO 3 concentration in CB was also higher than that in RS, especially in South Asia (Fig. S6(e)).
We summarized gaseous surface concentrations from six sites for comparison with the model simulations.The observed and simulated mean gaseous surface concentrations were shown in Table 2 and the corresponding normalized mean bias (NMB) were presented in Table S1.The comparison results indicated that both experiments overestimated the mean O 3 concentrations at Lhasha (NMBs are equal to 20.9 in CB and 52.7 in RS) and Kanpur (NMBs are equal to 56 in CB and 61.3 in RS), with the CB experiment exhibiting higher performance, whereas the O 3 concentration was underestimated at Namco, with the RS experiment exhibiting higher performance (NMB = -13.4).Our daily O 3 concentration measurement at Namco also illustrated the aforementioned finding (Fig. S7).The CB experiment showed higher performance for the surface NO 2 concentration at Lhasa, Aksu, and Lanzhou, whereas RS experiment revealed a closer agreement with observations at Kanpur.For the surface SO 2 concentration, the two experiments presented underestimation at Aksu and Lanzhou but overestimation at other sites.At Lhasha, the SO 2 concentration was clearly overestimated by 6.8 and 5.2 µg m -3 in CB and RS simulations, respectively.The simulation biases in NO 2 and SO 2 may partly be attributed to the deviation in modeling meteorological variables.In this study, the simulation underestimation of the surface wind speed (Fig. 4(c)) may not be beneficial for the dilution of pollutants and resulted in higher concentrations for gaseous species.

Aerosol Compounds
The CB and RS experiments showed a similar spatial distribution of the annual mean surface PM 10 and PM 2.5 concentration (Fig. 5).A higher aerosol concentration was observed in South Asia whereas a lower aerosol concentration was observed over the TP.The PM 10 concentration was higher in the RS experiment than in the CB experiment, particularly in South Asia, with the difference being up to 33 µg m -3 (Fig. 5(b)).This is partly due to the difference in aerosol compounds such as SIAs produced by gaseous compounds, and partly due to different distribution of the calculated concentrations between bins and modes.In the MADE/SORGAM aerosol scheme, dust emissions were mainly distributed in the fine fraction, and the smaller size of aerosol leads to a lower sedimentation rate and a higher aerosol concentration (Wei et al., 2011).Surface SO 4 , NO 3 , and NH 4 concentrations for both simulations are presented in Figs.6(a)-6(c).The annual mean SO 4 concentration was generally lower in CB than that in RS (Fig. 6(a)).As discussed previously, one of the most likely reasons of this difference was that the CBMZ gas-phase scheme in the CB experiment did not contain the aqueous-phase oxidation of SO 2 by H 2 O 2 .In the vertical profile, the SO 4 concentration in CB is lower than that in RS (Fig. S8(a)).The atmospheric NO 3 concentration is mainly determined by its precursor HNO 3 and the equilibrium between HNO 3 and nitrate.As shown in Fig. 5(b), these two experiments showed a quite similar distribution, and the differences were mainly appeared in regions with higher relative humidity (such as the southern TP and its south slope, Fig. 6(b)).Because of higher relative humidity, the gas-particle partitioning process from HNO 3 to NH 4 NO 3 was more efficient in the MADE/SORGAM scheme than in MOSAIC scheme.Thus, the Mozurkewich (1993) approach available in the MADE/SORGAM scheme depicted higher NO 3 concentrations over the southern TP and its south slope than did the Zaveri (2008) method available in the MOSAIC scheme (Fig. 6(c)).Higher NO 3 concentrations in the CB experiment also appeared in the vertical profile in the south slope of the TP.The CB experiment exhibited a lower annual mean NH 4 concentration than did the RS experiment (Fig. 6(d)) because of the underestimation of sulfate and nitrate concentrations in the MOSAIC aerosol module.The gas-phase HNO 3 and NH 3 can convert to particulate NH 4 NO 3 .Finally, the annual average concentrations of OC and BC are shown in Figs.6(e) and 6(f), respectively.Differences between the two experiments were 1.0 µg m -3 for OC and 1.25 µg m -3 for BC, with concentrations being always higher in the CB experiment.Higher BC concentration also appears in the vertical profile (Fig. S8(d)).Because of the same emission source, vertical dispersion, and dry deposition equation, discrepancies between the CB and RS experiments could be due to differences in aerosol dynamics and aerosol concentrations calculations between bins and modes; for example, the MOSAIC scheme in this study divides aerosols into four particle ranges, whereas the MADE/SORGAM scheme has only three modes.
A comparison of observed and simulated mean aerosol and its compounds at several sites is presented in Table 3 and the corresponding NMBs were presented in Table S2.The model showed underestimation for the aerosol concentration at Aksu and Lanzhou but overestimation at Lhasa.The CB and RS experiments showed simulation biases of 7.4 (NMB = 26.6)and 13.8 µg m -3 (NMB = 49.6) in the PM 2.5 concentration at Lhasa, respectively.For aerosol compounds at three sites in the TP (Lhasa, Namco, and Qomolangma), the CB experiment showed a closer agreement with observations.Over the TP, the RS configuration generally displayed a higher positive bias for surface SO 4 , NO 3 , and NH 4 concentrations and a higher negative deviation for surface BC and OC concentrations than did the CB configuration.Both the experiments underestimated the concentrations of aerosol compounds (SO 4 , NO 3 , and NH 4 ) at Lanzhou where the concentration of pollutants was higher.

Optical Properties
Fig. 7 presents the spatial distribution of the annual mean simulated aerosol optical depth at a wavelength of 550 nm (AOD550).The highest modeled AOD550 values were found in South Asia, whereas TP showed lower values in both experiments.Differences between the two simulations were up to 0.24 in South Asia, where the RS experiment revealed a higher amount of fine particle of aerosols than did the CB experiment (Fig. 5(b)).This finding suggests the crucial role of fine particles in AOD550 estimation, which have higher scatter efficiency than coarse particles (Seinfeld et al., 1998).In the TP, differences in the magnitude between the two experiments were less remarkable, with the values ranging from -0.03 to 0.03.Fig. 8 shows the time series of simulated and observed daily average AOD550 values at 15 AERONET sites, and the corresponding statistics are presented in Table S3.The two experiments efficiently represented the temporal variability of observed AOD550 values with the correlation ranging from 0.45 (RS) to 0.50 (CB).The RS experiment overestimated the observed values (mean NMB = 25.7),whereas the CB experiment underestimated the observed values (mean NMB = -7.4).This was primarily because most AERONET stations are located in Southern Asia and the RS experiment showed overestimation at these sites; that is, the simulated AOD550 value showed biases of -0.06 and 0.09 in the CB and RS experiment, respectively.
The considerable simulation differences in AOD550 values between the CB and RS experiments may partially be due to differences in the aerosol composition.A previous study indicated that fine (NH 4 ) 2 SO 4 and NH 4 NO 3 can scatter light more efficiently (Seinfeld et al., 1998).As evident in Fig. 5, the RS simulation showed higher nitrate and sulfate concentrations than did CB experiment in north India, whereas the two experiments presented similar magnitudes and spatial patterns over the TP.Thus, we selected Kanpur (located in north India) and Namco (located in the TP) to specifically analyze the connection between SIA differences and AOD550.Fig. 9 shows the scatterplots of daily differences in AOD550 values between the two experiments versus the corresponding differences in daily fine nitrate and sulfate concentrations, calculated in 2013.The correlations between SIA differences and AOD550 were beyond 0.6 in Kanpur and around 0.5 in Namco.This suggests that the two SIA compounds were relevant in AOD550 estimations, and higher surface sulfate and nitrate concentrations might lead to higher AOD550 values in RS than in CB, particularly in South Asia.Particle size is another crucial aspect affecting the simulation of AODs.The angstrom exponent at the interval of 440-675 nm, an indicator of particle distribution, was analyzed together with AODs, where values ranging from 1 to 2 suggest that small aerosols dominate in the accumulation mode, whereas values close to zero represent the existence of coarse particles (Eck et al., 1999).Specifically, the majority of angstrom exponent values were lower than 1 at Kanpur (Figs. 10(a) and 10(b)), indicating to dominance of coarse-mode aerosols in this region.The overestimation of AOD550 values in the RS experiment may be due to the distribution of dust emissions mainly in the fine fraction in the MADE/SORGAM scheme, and the fine particles had a higher scattering efficiency.By contrast, at the Namco station, which had a high angstrom exponent value, there was a better agreement with the measured AOD550 value in the CB experiment than in the RS experiment (Figs.10(c) and 10(d)).This is mainly because the MOSAIC scheme in CB divides aerosols into four particle ranges (MADE/SORGAM has only three modules), which can adequately represent fine particles at Namco.Taken together, deficiencies in AODs between the simulation and observation increased for the low angstrom exponent value and the CB simulation performed better for AODs in the TP.

CONCLUSIONS
In this study, two annual WRF-Chem modeling configurations were performed to investigate the effect of different chemical mechanisms on the reproduction of aerosol concentrations and aerosol optical properties over the TP and surrounding regions.Simulated results were compared with observations from CMDN, CRU, ERA-Interim, AERONET, and in situ stations collected from other papers.The results indicated that WRF-Chem could capture the spatial and temporal distribution of the surface temperature, relative humidity, and wind over the study area.The CBMZ gas-phase mechanism demonstrated lower O 3 and higher NO 2 and HNO 3 concentrations than did the RADM2, and this is because of more efficient O 3 -NO titration and a higher reaction rate constant for NO 2 + OH + M → HNO 3 in CBMZ.Lower SO 2 and higher H 2 O 2 surface concentration were observed in the RS experiment, because SO 2 could easily form particulate sulfate through cloud oxidation in RADM2.
Surface PM 2.5 and PM 10 concentrations obtained from CB and RS configurations presented a similar spatial pattern, and the RS experiment showed a higher aerosol concentration.This is partly due to the difference in aerosol compounds, such as SIAs produced by gaseous compounds.In particular, the annual mean SO 4 , NO 3 and NH 4 concentrations were higher in RS than in CB, and BC and OC concentrations were generally lower in RS.Additionally, different distribution of the calculated concentrations between bins and modes contributed to the discrepancy in aerosol concentrations.
The CB experiment revealed a closer agreement with the observed values of aerosol and its compounds in the TP.The model sufficiently captured the spatial and temporal pattern of AOD550 values and had a correlation ranging from 0.45 (RS) to 0.50 (CB) with observations at 15 AERONET sites.In the two simulation configurations, AOD550 showed potential positive correlation with SIA (e.g., sulfate and nitrate).Particle size was another crucial aspect in modeling AODs.At Namco, the CB experiment more satisfactorily represented AODs than did the RS, because the MOSAIC scheme divides aerosols into four particle bins that can better reflect the particle size distribution over the TP, whereas MADE/SORGAM has only three bins.
The results of this study suggest that the selection of chemical mechanisms is a key aspect of model simulation, even when the online-coupled WRF-Chem model is concerned.The CBMZ gas-phase chemical mechanism and MOSAIC aerosol scheme could satisfactorily reconstruct the reconstruction the concentrations and optical depth of aerosols over the TP.Nonetheless, coarse particles were still poorly represented by the chosen aerosol schemes.Therefore, a revision of the treatment concerning the coarse aerosol size distribution is highly recommended in the future, such as the MOSAIC aerosol scheme with eight sectional bins.

Fig. 1 .
Fig. 1.Model domain and topographic field (m); the black and red dots represent the CMDN and AERONET stations, respectively.
Fig. 2. Mean surface air temperature (a-c), surface relative humidity (d-f) and wind at 500 hPa for CB (left), RS (center) simulation and CRU(ERA-Interim) data (right) in 2013.

Fig. 3 .
Fig. 3. Daily mean variation of measured and simulated (a) T2, (b) RH2, and (c) U10 with box plot at meteorological stations in study domain in 2013.

Fig. 5 .
Fig. 5. PM 10 (a) and PM 2.5 (b) yearly mean concentrations at the ground for CB (left), RS (center) experiments and the discrepancy between CB and RS (right).

Fig. 9 .
Fig. 9. Scatter plot diagram of AOD550 daily differences versus SO 4 (left) and NO 3 (right) daily differences at Kanpur (upper) and Namco (under) AERONET stations in 2013.Differences are calculated as CB-RS.
This work was supported by the Key Research Program of the Chinese Academy of Sciences (KJZD-EW-G03-04), the National Natural Science Foundation of China (41630754, 91644225, 41701074), the Open Program (SKLCS-OP-2017-02) from State Key Laboratory of Cryospheric Science, and the China Postdoctoral Science Foundation (2016M602896).The computer resources were supported by the Supercomputing Center, Big Data Center of Cold and Arid Region Environment and Engineering Research Institute, Chinese Academy of Sciences and we are also grateful to Guohui Zhao for his help of installing software.

Table 3 .
Observed and simulated aerosol and its compounds' concentration (µg m -3