Impacts of heterogeneous uptake of dinitrogen pentoxide and chlorine activation on ozone and reactive nitrogen partitioning: Improvement and application of WRF-Chem model in southern China

Abstract. The uptake of dinitrogen pentoxide (N2O5) on aerosol surfaces and the subsequent production of nitryl chloride (ClNO2) can have a significant impact on the oxidising capability and thus on secondary pollutants such as ozone. The range of such an impact, however, has not been well quantified in different geographical regions. In this study, we applied the Weather Research and Forecasting coupled with Chemistry (WRF-Chem) model to investigate the impact of the N2O5 uptake processes in the Hong Kong–Pearl River Delta (HK–PRD) region, where the highest ever reported N2O5 and ClNO2 concentrations were observed in our recent field study. We first incorporated into the WRF-Chem an aerosol thermodynamics model (ISORROPIA II), recent parameterisations for N2O5 heterogeneous uptake and ClNO2 production and gas-phase chlorine chemistry. The revised model was then used to simulate the spatiotemporal distribution of N2O5 and ClNO2 over the HK–PRD region and the impact of N2O5 uptake and Cl activation on ozone and reactive nitrogen in the planetary boundary layer (PBL). The updated model can generally capture the temporal variation of N2O5 and ClNO2 observed at a mountaintop site in Hong Kong, but it overestimates N2O5 uptake and ClNO2 production. The model results suggest that under average conditions, elevated levels of ClNO2 (> 0.25 ppb within the PBL) are present in the south-western PRD, with the highest values (> 1.00 ppb) predicted near the ground surface (0–200 m above ground level; a.g.l.). In contrast, during the night when very high levels of ClNO2 and N2O5 were measured in well-processed plumes from the PRD, ClNO2 is mostly concentrated within the residual layer ( ∼  300 m a.g.l.). The addition of N2O5 heterogeneous uptake and Cl activation reduces the NO and NO2 levels by as much as 1.93 ppb ( ∼  7.4 %) and 4.73 ppb ( ∼  16.2 %), respectively, and it increases the total nitrate and ozone concentrations by up to 13.45 µg m−3 ( ∼  57.4 %) and 7.23 ppb ( ∼  16.3 %), respectively, in the PBL. Sensitivity tests show that the simulated chloride and ClNO2 concentrations are highly sensitive to chlorine emission. Our study suggests the need to measure the vertical profiles of N2O5 ∕ ClNO2 under various meteorological conditions, to consider the chemistry of N2O5 ∕ ClNO2 in the chemical transport model and to develop an updated chlorine emission inventory over China.


Introduction
Dinitrogen pentoxide (N 2 O 5 ) is mostly produced by chemical reactions involving ozone (O 3 ) and nitrogen dioxide (NO 2 ).
N 2 O 5 (g) + H 2 O (aq) −→ 2 HNO 3 (aq) (R3) N 2 O 5 (g) + HCl (aq) −→ HNO 3 (aq) + ClNO 2 (g) (R4) The above processes affect the fate and composition of the total reactive nitrogen (NO y ), which is the sum of NO, NO 2 , HNO 3 (g), 2 × N 2 O 5 , NO 3 , ClNO 2 , PAN, HONO, HNO 4 , aerosol nitrate and various organic nitrates.The hydrolysis of N 2 O 5 is a major loss pathway for NO x (= NO + NO 2 ) at night, reducing the amount of NO x for daytime photochemistry in the following day, while producing nitrate aerosol contributing to secondary aerosol (Brown and Stutz, 2012).When ClNO 2 is produced, it serves as a reservoir for reactive nitrogen at night, and it is photolysed to recycle NO 2 and release highly reactive chlorine radicals (Cl activation), both of which can significantly affect the daytime photochemistry, such as O 3 formation via reactions with volatile organic compounds (VOCs; Atkinson, 2000;Thornton et al., 2010;Riedel et al., 2014).
The critical parameters required to determine the impacts of the N 2 O 5 uptake processes are the rate constant of Reaction (R5), k 5 , and the yield of ClNO 2 , φ. k 5 can be calculated from Eq. (1) by treating the N 2 O 5 heterogeneous uptake reaction as a first-order reaction (Chang et al., 2011): where V N 2 O 5 denotes the mean molecular velocity of N 2 O 5 , S aer is the aerosol surface area density and γ represents the heterogeneous uptake coefficient of N 2 O 5 which is the possibility that a collision of an N 2 O 5 molecule with a particle will lead to uptake and a chemical reaction (Sarwar et al., 2012).
V N 2 O 5 and S aer are relatively well determined; therefore, the treatments of γ and φ are crucial for the prediction of the impacts of N 2 O 5 uptake and Cl activation.In terms of γ , a fixed value of 0.1 was first proposed (Dentener and Crutzen, 1993).
Highly variable γ values were then derived from observational campaigns, which showed differing values for inland and marine aerosols (e.g. Brown et al., 2006Brown et al., , 2009;;Osthoff et al., 2008).Later, laboratory and modelling studies considered the dependence of γ on the aerosol species/compositions (sea salt, black carbon, sulfate, nitrate, chloride, organic matter and water), temperature and/or relative humidity (Evans and Jacob, 2005 and the reference therein; Davis et al., 2008;Anttila et al., 2006;Riemer et al., 2009;Bertram and Thornton, 2009).Several parameterisations have been proposed for the yield of ClNO 2 .Simon et al. (2010) applied a constant value of 0.75 for the fraction of N 2 O 5 involved in the production of ClNO 2 .More detailed parameterisations of φ considering the effects of aerosol compositions were proposed by Roberts et al. (2009) and by Bertram and Thornton (2009).
Several studies examined the impacts of N 2 O 5 uptake or ClNO 2 production with the use of the chemical transport model.Dentener and Crutzen (1993) evaluated the impacts of the N 2 O 5 uptake on tropospheric aerosol by using a constant γ (0.1), and their results showed a substantial decrease of NO x (50 %) and a slight reduction of O 3 (9 %) globally.Riemer et al. (2003) utilised a weighting factor on a constant γ (0.02) to study the N 2 O 5 heterogeneous hydrolysis, and their research showed small impacts on O 3 and remarkable influences on nitrate formation in Europe.By applying parameterisations of γ for various aerosol compositions and meteorological conditions, Evans and Jacob (2005) showed higher levels of O 3 and NO x compared to simulations with a γ value of 0.1.Lowe et al. (2015) and Archer-Nicholls et al. (2014) incorporated the heterogeneous uptake of N 2 O 5 on particles into the MOSAIC aerosol module in the Weather Research and Forecasting coupled with Chemistry (WRF-Chem) model based on the methods suggested by Bertram and Thornton (2009), Anttila et al. (2006) and Riemer et al. (2009).Their results suggested that N 2 O 5 uptake suppressed VOC oxidation (by OH and NO 3 ) by a factor of 1.5 and significantly enhanced nitrate formation during nighttime (an increase from 3.5 to 4.6 µg kg −1 ) over north-western Europe.Simon et al. (2009) used a gas-phase reaction to represent the heterogeneous production process of ClNO 2 , and their study predicted modest increases in O 3 due to the ClNO 2 production in Houston.Sarwar et al. (2012) implemented the heterogeneous production of ClNO 2 based on the parameterisation proposed by Bertram and Thornton (2009) and additional gas-phase chlorine reactions in CMAQ, and they examined the impacts of Cl activation due to ClNO 2 production and Cl chemistry on air quality.Their results showed that ClNO 2 production reduced the total nitrate level (up to 0.8-2.0 µg m −3 or 11-21 %) and had modest impacts on the 8 h O 3 level (up to 1-2 ppb or 3-4 %) in the United States.Sarwar et al. (2014) expanded the study region used in Sarwar et al. (2012) to the entire Northern Hemisphere, and they suggested that ClNO 2 production had remarkable impacts on the air quality in China and western Europe, with enhancements of the 8 h O 3 level up to 7.0 ppb.Most previous studies focused on investigating the effects of N 2 O 5 uptake or ClNO 2 production in North America and Europe; however, little is known about Asia.The only study that covered Asia was performed by Sarwar et al. (2014); it used a coarse model resolution (> 100 km) and considered only biomass burning and sea salt as the source of chloride.
Previous studies in Asia (Hong Kong) have revealed the existence and significance of ClNO 2 in this region (Wang et al., 2014;Tham et al., 2014;Xue et al., 2015).In a recent field study, we observed the highest ever reported mixing ratios of N 2 O 5 (1 min average value up to 7.7 ppbv) and ClNO 2 (4.7 ppbv) at a mountaintop site (957 m above sea level) in Hong Kong (Brown et al., 2016;Wang et al., 2016).This result indicates rapid production of N 2 O 5 and ClNO 2 in the Hong Kong-Pearl River delta (HK-PRD) region, which has long suffered O 3 and NO x pollution (Wang et al., 2009).Meteorological analysis and chemical data revealed highly inhomogeneous dynamic and chemical processes and considerable impacts of the ClNO 2 chemistry on the radical budget (up to 77 % increase of OH) and daytime O 3 production (up to 41 %) in the high ClNO 2 air mass as it transported to downwind locations above the ocean (Wang et al., 2016).It was also suggested in that study that other locations downwind of major urban areas under prevailing conditions may experience more frequent events with high levels of ClNO 2 / N 2 O 5 than the site in Hong Kong.It is therefore of great interest to investigate the regional distribution of N 2 O 5 / ClNO 2 and the impact of N 2 O 5 uptake and subsequent chemistry.
This study was conducted to investigate (1) the spatial (horizontal and vertical) distribution of the N 2 O 5 and ClNO 2 concentrations in the HK-PRD region and (2) the spatial extent of the impact of N 2 O 5 uptake processes on the formation of O 3 and the partitioning of reactive nitrogen in this region.The latest version of a widely used aerosol thermodynamics model, ISORROPIA II (Fountoukis and Nenes, 2007), was incorporated into the MADE/VBS aerosol model.ISOR-ROPIA II has the ability to simulate the equilibrium between hydrogen chloride (HCl) and chloride, which is critical for the simulation of N 2 O 5 heterogeneous uptake and Cl activation.However, this capacity is not considered in the current MADE/VBS model in WRF-Chem (Grell et al., 2005;Ahmadov et al., 2012).Up-to-date parameterisation for N 2 O 5 uptake and ClNO 2 production and Cl-radical-initiated chemistry were implemented into WRF-Chem.The revised WRF-Chem was then applied to southern China to investigate the spatial distribution of N 2 O 5 and ClNO 2 and the impacts of these processes on O 3 and NO y .We start with a description of the data used to run and validate the simulations, the amendments to the WRF-Chem model and the model setup in Sect. 2. In Sect.3, we show the performance of the WRF-Chem model in the simulation of several air pollutants without N 2 O 5 uptake processes and the simulation results of N 2 O 5 and ClNO 2 with N 2 O 5 uptake and Cl activation processes; we then evaluate the impacts of N 2 O 5 uptake and Cl activation on NO x , total nitrate and O 3 and NO y partitioning, and we test the sensitivity of the ClNO 2 concentration to chlorine emissions.A summary is given in Sect. 4.

Methodology
2.1 Data 2.1.1Field measurement data N 2 O 5 and ClNO 2 concentrations were measured at Tai Mo Shan (TMS) in Hong Kong with a chemical ionisation mass spectrometer between 15 November and 5 December 2013 (refer to Wang et al., 2016, for more details).The measurements were made on a mountaintop in the south-eastern PRD at an altitude of 957 m.Other major air pollutants, including PM 2.5 , NO 2 and O 3 , were also measured at the TMS site and at 11 general (non-roadside) monitoring stations of the Hong Kong Environmental Protection Department (HKEPD; available at http://epic.epd.gov.hk/EPICDI/air/station/).Hourly measurement data were used to validate the performance of the WRF-Chem simulations.

Emission data
Four sets of anthropogenic emission inventories (EIs) covering different areas were adopted in this study.For mainland China, we used the Multi-resolution Emission Inventory for China (MEIC; available at http://meicmodel.org), developed by Tsinghua University for the year 2010.For the PRD, the anthropogenic EI developed by the Southern China University of Technology was applied.The anthropogenic EI developed by HKEPD was used over Hong Kong.INTEX-B EI (Zhang et al., 2009) was adopted for other Asian regions.Readers are referred to Zhang et al. (2016) for the details of these anthropogenic EIs.For natural emissions, the biogenic emission parameterisation proposed by Guenther et al. (1994), the dust emission parameterisation proposed by Shaw et al. (2008) and the sea salt emission parameterisation proposed by Gong et al. (2002) were adopted in this study.The concentrations of sodium, chloride, calcium, magnesium and potassium in dust and sea salt follow those suggested by Millero (1996) and Wedepohl (1995), as shown in Table S1 in the Supplement.
Chlorine emissions are not included in most EIs, but they are critical for the simulation of N 2 O 5 uptake and Cl activation.In this study, the Reactive Chlorine Emission Inventory (RCEI; Keene et al., 1999  and anthropogenic activities (gas phase; e.g.coal combustion).Global chlorine emissions from biomass burning and anthropogenic activities are estimated to be ∼ 6.3 and ∼ 6.6 Tg Cl yr −1 , respectively.The RCEI inventory is the only available chlorine EI that currently covers China, and it is subject to some, probably large, uncertainties for representing Cl emissions in the HK-PRD region due to its low spatial resolution and the fact that it was compiled for the year 1990.Coal consumption and SO 2 emissions from coalfired power plants in China increased by 479 and 56 % from 1990 to 2010, respectively (Liu et al., 2015); thus it is expected that chlorine emissions from coal combustion, which form a large proportion of anthropogenic chlorine (Keene et al., 1999), also increased significantly over that period.We conducted two sensitivity simulations by adjusting the chlorine emissions to test the dependence of the chloride and ClNO 2 concentrations on the varied chlorine emissions; the results are shown in Section 3.5.

Meteorological data
Three-hourly meteorological measurements, including atmospheric pressure, temperature, relative humidity, wind direction and wind speed, at ∼ 2500 surface meteorological stations and 12-hourly data at ∼ 250 sounding stations, were obtained from the China Meteorological Agency (CMA) and Hong Kong Observatory and were adopted in 4-D data assimilation to improve the model performance of the meteorological fields using observational nudging techniques (Zhang et al., 2016).The FNL Operational Global Analysis dataset provided by the National Centers for Environmental Prediction (available at http://rda.ucar.edu/datasets/ds083.2/) was used for analysis nudging.Observational and analytical nudging techniques have been shown to improve the performance of meteorological simulation in both northern China (Zhang et al., 2015) and southern China (Zhang et al., 2016).Hourly datasets from ∼ 500 surface meteorological observation stations obtained from the CMA were used to validate the meteorological simulations.

Incorporation of ISORROPIA II
ISORROPIA II, an aerosol thermodynamics model developed by Nenes et al. (1998) and Fountoukis and Nenes (2007), was incorporated to replace the aerosol thermodynamics module in the MADE/VBS aerosol model in the original WRF-Chem v3.5 so as to extend the capacity of simulation of the equilibrium between PM 2.5 compositions and their corresponding gaseous species.The MADE/VBS model adopts a volatility basis set model to simulate secondary organic aerosol (SOA) formation, and it provides improved simulations of SOA compared to the traditional MADE/SORGAM model (Ahmadov et al., 2012).The current MADE/VBS model only estimates the thermodynamic equilibrium between SO 2− 4 , NO − 3 , NH + 4 , H 2 O and corresponding gases, whereas ISORROPIA II simulates the equilibrium between SO 2− 4 , NO − 3 , NH + 4 , H 2 O, Na + , Cl − , Ca 2+ , Mg 2+ , K + and associated gases.

N 2 O 5 heterogeneous uptake, ClNO 2 production and Cl gaseous reaction
We adopted the parameterisations of N 2 O 5 heterogeneous uptake and ClNO 2 production suggested by Bertram and Thornton (2009).According to the parameterisations, the N 2 O 5 heterogeneous uptake coefficient, γ , can be calculated with the following equation: ), where ] are the molarities of liquid water, nitrate and chloride in aerosol volume.The yield of ClNO 2 , φ, can be calculated with the following equation: The loss of N 2 O 5 and the production of nitrate and ClNO 2 can be predicted with Eqs. ( 1)-( 3).The produced ClNO 2 is then photolysed, releasing a Cl atom, which further oxidises VOCs like an OH radical.The Cl-initiated gaseous chemistry used in this study was originally designed for the CB05 mechanism by Sarwar et al. (2012) and was modified for the RACM_ESRL mechanism (detail reactions are shown in Table S2).RACM_ESRL mechanism is the updated Regional Atmospheric Chemistry Mechanism in WRF-Chem based on the original version in Stockwell et al. (1997).The photolysis rates of Cl 2 , HOCl, ClNO 2 and formyl chloride (FMCl) were calculated with the absorption cross section and quantum yield obtained from Atkinson et al. (2007) and Atkinson et al. (2008).We implemented the N 2 O 5 heterogeneous uptake, the ClNO 2 production and the Cl-initiated reactions into the MADE/VBS aerosol model, RACM_ESRL gas-phase mechanism and Madronich photolysis model (Madronich, 1987) in the "RACM_SOA_VBS_KPP" chemistry option in WRF-Chem v3.5.

Model configuration
The model configurations of WRF-Chem used in this study are shown in Table 1.We used the Noah model to simulate the land surface process, the YSU module to simu- late the planetary boundary layer (PBL) processes, the Purdue Lin scheme to predict the microphysics, the Grell 3-D ensemble module to simulate the cumulus, the RRTMG model to predict shortwave and longwave radiation and the RACM_ESRL, MADE/VBS and Madronich modules to simulate gas-phase chemistry, aerosol processes and photolysis.Model simulations were conducted in four domains covering eastern Asia, southern China, the PRD and Hong Kong, with spatial resolutions of 27, 9, 3 and 1 km, respectively (see Fig. 1a).High grid resolutions were adopted in this study to capture the extremely inhomogeneous terrain, with land and sea, mountain and plains and urban and forested areas, as shown in the terrain map of domain 2 (southern China) in Fig. 1b.The red dotted line in Fig. 1b represents the vertical cross-section domain that intercepts the most polluted part of the PRD and follows the prevailing (north-east) wind direction.The vertical domain is used to illustrate the vertical distribution of the N 2 O 5 and ClNO 2 concentrations and the impacts of N 2 O 5 uptake processes in southern China.Thirty vertical model layers were adopted, of which eight layers are below 1000m (approximately the height of PBL at noon) to provide more detailed information within the PBL.

Simulation cases
Three simulation cases, shown in Table 2, were conducted from 15 November to 5 December 2013, during which ClNO 2 and N 2 O 5 levels were measured at the TMS site.All simulations used ISORROPIA II as the aerosol thermodynamics module.Note that the base case did not include N 2 O 5 heterogeneous uptake (or ClNO 2 chemistry).The HET+Cl case included the complete N 2 O 5 uptake and Cl activation processes, i.e.N 2 O 5 loss on aerosol, ClNO 2 production and gaseous chlorine reactions.Differences in chemical concentrations between the base and HET+Cl cases, i.e.HET+Clbase, represent the impacts of N 2 O 5 uptake and Cl activation.To estimate the relative contribution of N 2 O 5 uptake vs. Cl activation to O 3 and NO y partitioning, the HET case was also conducted, which included N 2 O 5 uptake but not ClNO 2 production (i.e.ClNO 2 yield, φ, was set to 0) and therefore producing only nitrate from N 2 O 5 uptake.The changes from the base case to the HET case (HET-base) represent the impacts of N 2 O 5 heterogeneous uptake, whereas the changes from the HET case to the HET+Cl case (HET+Cl-HET), represent the impacts of Cl activation.

Model performance of WRF-Chem without N 2 O 5 uptake and Cl activation
The meteorological simulation determines the simulations of the transport of the air pollutants and therefore is crucial to the simulations of the spatial distributions of the atmospheric chemical species and their impacts.The performance of the meteorological module during the study period has been validated in Wang et al. (2016), which showed that atmospheric flow and other meteorological parameters were satisfactorily   simulated.The reader is referred to Wang et al. (2016) for further details.
The chemical simulation results of WRF-Chem without N 2 O 5 heterogeneous uptake and Cl activation, i.e. the base case, were validated against hourly observations of several major air pollutants measured at 11 HKEPD stations and at the TMS site.PM 2.5 , NO 2 and O 3 were selected as the validation species because they act as the reaction surface (PM 2.5 ) or precursors (NO 2 and O 3 ) for N 2 O 5 and ClNO 2 production.As shown in Table 3, the validation results for HKEPD stations indicate that the base case simulated the major air pollutants reasonably well in this region but overestimated PM 2.5, slightly underestimated NO 2 and underestimated O 3 .The base case also generally reproduced the observed temporal variations of PM 2.5, NO 2 and O 3 at the HKEPD stations and the TMS site (Fig. S1a, b, c in the Supplement), and simulated the level of aerosol surface area density and particulate nitrate at TMS (Fig. S1d, e).It should be noted that the technique for measuring NO 2 by the HKEPD, which is similar to that used in the regular air monitoring networks in North America and Europe, employs catalytic conversion that overmeasures NO 2 (e.g.Xu et al., 2013).The discrepancy between the simulated and observed major air pollutants in this area is expected to affect the simulations of N 2 O 5 and ClNO 2 , which will be discussed in Sect.3.2.1.The model performance of major air pollutants of the base case is within the acceptable range and is similar to our previous applications of WRF-Chem (Zhang et al., 2015(Zhang et al., , 2016) ) and other WRF-Chem model studies (e.g.Li et al., 2011).
The simulated fine chloride concentrations in the base case were compared with observations from several campaigns, as shown in Table 4. Tan et al. (2009) reported average concentrations of 1.19 and 8.37 µg m −3 at an urban site in Guangzhou (GZ) in the PRD on normal days and hazy days in winter, respectively; in comparison, the base case simulated an average level of 2.51 µg m −3 at that location.Tao et al. (2014) reported an average level of 3.30 µg m −3 in winter at the station of the South China Institute of Environmental Science (SCIES) in Guangzhou; the base case predicted 2.13 µg m −3 at this location.At the Tung Chung (TC) site in Hong Kong, we had previously measured an average level of 1.10 µg m −3 of chloride in PM 2.5 in late autumn and early winter, while the base case simulated 0.32 µg m −3 .At the TMS site, an average level 0.37 µg m −3 of chloride was observed during the campaign (which was also the simulation period of this study), while the base case predicted 0.14 µg m −3 .The simulated chloride level in the base case was of the same order of magnitude with observations over  the PRD, but it still underpredicted the observed chloride level due to the expected underestimates of chlorine sources in the EI we applied (see Sect. 2.1.2).
The simulated ratio of aerosol surface area to volume ranges between 0.5 × 10 7 and 2.5 × 10 7 m −1 (Fig. S2), which is comparable with the value (2.67 × 10 7 m −1 ) used in the parameterisation of N 2 O 5 uptake coefficient on aerosol proposed by Bertram and Thornton (2009).
Overall, the validations of the meteorological and chemical simulations suggest that the model is capable of capturing the general characteristics of air flow and key atmospheric chemical processes and hence can provide a basis for further simulation of the distributions of the N 2 O 5 and ClNO 2 concentrations and the impacts of N 2 O 5 uptake and Cl activation on NO y partitioning and O 3 production.in lower production of N 2 O 5 ; the simulated PM 2.5 concentrations are higher than the observed values, which would lead to an overestimate of N 2 O 5 heterogeneous loss.The observation-derived N 2 O 5 uptake coefficients at the TMS site (Brown et al., 2016) varied from 0.004 to 0.029, with an average value of 0.014, whilst the simulated uptake coefficients ranged from 0.008 to 0.031, with an average of 0.019, which suggests that the HET+Cl simulation generally overestimates N 2 O 5 uptake coefficients, which causes further overestimation of the loss of N 2 O 5 .The reactive uptake coefficient could be overestimated because the parameterisation used in this study does not consider the inhibition of organic coating to the uptake coefficient (Bertram and Thornton, 2009).The overestimated loss of N 2 O 5 on aerosol inherently overestimated the production of ClNO 2 .The parameterisations used in this study are likely to overestimate the ClNO 2 yield (Kim et al., 2014;Ryder et al., 2015), which would further overestimate the production of ClNO 2 .In addition, the overpredicted ClNO 2 could also be due to the ignorance of possible ClNO 2 sinks (e.g.Roberts et al., 2008).
Discrepancies between the measured and simulated N 2 O 5 and ClNO 2 levels have also been reported in previous model studies.Lowe et al. (2015) used the same parameterisations for N 2 O 5 uptake that we applied in our study and showed slightly higher average simulated N 2 O 5 values along two flight tracks but simulated N 2 O 5 in another flight a factor of 1-2 lower.They noted that the underestimated N 2 O 5 could be attributed to inaccuracies in the meteorological simulation.Sarwar et al. (2012) used the parameterisation for N 2 O 5 uptake proposed by Davis et al. (2008) and by Bertram and Thornton et al. (2009) and yielded a slightly higher simulated peak value of ClNO 2 than the observed value in field studies conducted at different times from the model simulations.The authors attributed the overestimate of ClNO 2 to the overestimated N 2 O 5 uptake in the parameterisations.Sarwar et al. (2014) predicted lower peak values of ClNO 2 than the observations and suggested that the underestimated ClNO 2 could be attributed to a relatively low model resolution (108 km).

Spatial distribution of average simulated N 2 O 5 and ClNO 2
Figure 3a and c show the average mixing ratios of N 2 O 5 and ClNO 2 during the night-time (18:00-07:00 LT) within the lowest 1000 m (the approximate height of the PBL at noon) in southern China in the HET+Cl case.Elevated levels of N 2 O 5 (> 0.10 ppb) and ClNO 2 (> 0.25 ppb) were predicted in the areas downwind of the PRD, as a result of the transport of pollutant enriched air masses towards the south-west of the PRD by the prevailing north-easterly winds.The areas with the highest simulated N 2 O 5 and ClNO 2 values did not cover the TMS site at which the highest ever reported N 2 O 5 and ClNO 2 values were observed (Brown et al., 2016;Wang et al., 2016), which supports our speculation that the locations downwind of the PRD under the dominant northeasterly winds may frequently have higher levels of ClNO 2 .The simulated yield of ClNO 2 during the night-time ranged from 0.1 to 0.7 within the PBL (Fig. S4), which is consistent with a previous observation study (0.1 to 0.65; Osthoff et al., 2008) and modelling study (0.0 to 0.9; Sarwar et al., 2012).The vertical distributions of N 2 O 5 and ClNO 2 in the vertical domain (as described in Sect.2.3.1) are shown in Fig. 3b and d.Elevated levels of N 2 O 5 (> 0.10 ppb) were predicted up to around 1000 m a.g.l., with the highest N 2 O 5 level (> 0.25 ppb) mostly between 400 and 800 m a.g.l., probably due to the reactions of NO 3 with NO and VOCs in the lowest several hundred metres over the urban area, as

Dynamic evolutions in cases with typical and extreme meteorological conditions
We examine the time evolution of the spatial distribution of ClNO 2 in two cases.In the typical case (the night of 1/2 December), southern China was dominated by consistent northeasterly winds which represented the average dynamic conditions during the study period, while in the extreme case (the night of 3/4 December), the air flow over the region abruptly changed.Note that in this extreme case, the highest ever reported ClNO 2 levels were observed at the TMS site, and the back trajectories and observations of chemical species pointed to the transport to the site of well-processed plumes from the PRD with enriched anthropogenic chloride and other pollutants (Wang et al., 2016).
In the typical case, consistent north-easterly winds controlled southern China throughout the night (Fig. 4).At the beginning of the night (18:00 local time; LT), ClNO 2 began to build up near the urban area (Fig. 4a) and near the surface (Fig. 4b); at midnight (00:00 LT), the air with an elevated level of ClNO 2 moved to coastal areas (Fig. 4c) and accumulated near the surface (Fig. 4d); at dawn (06:00 LT), the peak ClNO 2 level was predicted over the open sea (Fig. 4e), and was pumped up to higher altitudes, with the peak value near the surface (Fig. 4f), due to the higher boundary layer height over the ocean, as shown in Fig. S7.
In the extreme case, at the beginning of the evening (18:00 LT), southern China had unfavourable dispersion conditions over the land, including inconsistent wind directions and low wind speeds.The air pollutants emitted from the PRD slowly swirled over it, as shown in Fig. 5a, resulting in a longer "cooking" time for ClNO 2 production.The vertical distribution (Fig. 5b) shows that ClNO 2 -enriched air stretched from the ground up to 800 m a.g.l.The enhanced production of ClNO 2 is believed to be partially responsible for the highest ClNO 2 mixing ratios measured at the TMS site during that night.At midnight (00:00 LT), inconsistent wind directions were evident between land and sea areas: northerly winds dominated over the land area, while northeasterly winds dominated over the sea, leading to relatively slow motion of the ClNO 2 -enriched plume from the land towards the ocean (Fig. 5c).The vertical distribution (Fig. 5d) suggests that ClNO 2 built up within the residue layer.At dawn (06:00 LT), the north-easterly wind regained control over the land areas, and the air with the elevated level of ClNO 2 (> 2.00 ppb) was driven towards the ocean, as shown in Fig. 5e.The vertical distribution (Fig. 5f) shows that the peak ClNO 2 concentration was predicted to be in the residue layer at ∼ 300 m a.g.l.The changes of wind flow over the region during this night resulted in abnormal changes in the history of the air masses that reached the TMS site and led to the abrupt changes in the air pollutants concentrations obthere (see Wang et al., 2016, for details).From these results, it can be seen that the vertical distributions of ClNO 2 demonstrated distinct features in the two cases.To understand the underlying cause, it is of significance to measure the vertical profiles of ClNO 2 under various meteorological conditions.In addition, during the extreme event, the location with the highest predicted ClNO 2 (> 2.00 ppb) was not at the TMS site (> 1.00 ppb), but it was located in the western parts of the PRD (i.e. the cities of Jiangmen and Zhaoqing), which supports the contention that the ClNO 2 concentrations at other locations could be even higher than those observed at the TMS site (Wang et al., 2016).It would be of great interest to conduct measurements in the areas where the highest ClNO 2 concentrations are predicted.

Impacts of N 2 O 5 heterogeneous uptake and Cl
activation on NO x , total nitrate and O 3

Impacts on horizontal and vertical distributions of the species
Figure 6 shows the simulation results for the average NO, NO 2 , total nitrate and O 3 concentrations within the PBL (< 1000 m) during the entire simulation period in the base case, and the difference of the results between the HET+Cl and base cases in the horizontal domain.Relatively high levels of NO (up to 26.18 ppb; Fig. 6a), NO 2 (up to 29.18 ppb; Fig. 6c), total nitrate (up to 23.43 µg m −3 ; Fig. 6e) and O 3 (up to 44.50 ppb; Fig. 6g) were predicted in southern China in the base case.The spatial distributions of maxima of NO (Fig. S8a), NO 2 (Fig. S8c) and total nitrate (Fig. S8e) were similar to those of their daily averages, whereas the spatial distribution of 1 h maxima O 3 (higher in PRD and its downwind areas; Fig. S8g) was different from its daily average (higher in rural and remote region; Fig. 6g).With the prevailing north-easterly wind, the pollutants emitted from the PRD were transported towards the south-west, resulting in the most polluted regions being the PRD and its southwesterly downwind areas.After addition of the N 2 O 5 uptake and Cl activation processes, the NO (Fig. 6b) and NO 2 (Fig. 6d) levels were significantly decreased in the entire domain by up to 1.93 ppb (∼ 7.4 %) and 4.73 ppb (∼ 16.2 %), respectively.The regions with greater impacts on the NO and NO 2 due to the added processes were mostly urban and suburban areas with large emissions of NO x .A significant portion of NO x was transformed into total nitrate, which increased by as much as 13.45 µg m −3 (∼ 57.4 %) through the heterogeneous uptake of N 2 O 5 (see Fig. 6f).As can be seen in Fig. 6h, the N 2 O 5 uptake and Cl activation noticeably increased the O 3 levels across southern China, with a maximum increase up to 7.23 ppb (∼ 16.3 %).It is worth noting that in addition to the urban and suburban areas, the O 3 levels over the rural and coastal areas were also significantly affected by the added processes.The average impacts of N 2 O 5 uptake and Cl activation on daily maxima of NO, NO 2 , total nitrate and O 3 were shown in Fig. S8b, d, f and h, which had similar spatial patterns to the averaged results of the entire simulation period.The incorporation of N 2 O 5 uptake and Cl activation improved the model performance for daily O 3 and 1 h maximum O 3 (Table 3), but it did not improve the simulations of PM 2.5 and NO 2 , which could be due to various reasons, e.g.emission inventory.
Figure 7 shows the average simulated NO, NO 2 , total nitrate and O 3 values in the base case, and it shows the difference in the results between the base and HET+Cl cases in the domain.NO and NO 2 were concentrated within 800 m a.g.l.over the PRD and in downwind areas (see Fig. 7a and c).Total nitrate accumulated near the ground and stretched up to 800 m a.g.l.(Fig. 7e).Due to the titration effect of NO, relatively low average values of O 3 were simulated over urban areas (Fig. 7g).As shown in Fig. 7b and d, the N 2 O 5 uptake and Cl activation decreased the NO and NO 2 levels across the vertical domain, with the largest impacts seen in the near-surface layer (0-400 m a.g.l.) over the PRD.The lost NO and NO 2 were mostly transformed into total nitrate, which increased remarkably in the near-surface layer (Fig. 7f).The impacts of N 2 O 5 uptake and Cl activation on the O 3 level varied with altitude: the O 3 increased throughout the lowest 800 m with the largest enhancement near the ground, whereas it decreased above 1000 m a.g.l.(Fig. 7h).The changes in the O 3 concentrations were attributed to the combined effects of NO x loss due to N 2 O 5 uptake and Cl atom production due to Cl activation, both of which have non-linear impacts on O 3 production.The relative contribution of N 2 O 5 uptake vs. Cl activation on the NO, NO 2 , total nitrate and O 3 concentrations will be discussed in the following section.

Relative contribution of N 2 O 5 heterogeneous uptake vs. Cl activation
To understand the relative contribution of N 2 O 5 uptake and Cl activation, we conducted a sensitivity case (HET case as listed in Table 2) in which only nitrate was produced via N 2 O 5 uptake.The differences in the simulations between the base and HET cases represent the effects of N 2 O 5 uptake, while those between the HET and HET+Cl cases represent the effects of Cl activation.As shown in Fig. 8a and c, the mere consideration of nitrate production from N 2 O 5 uptake led to decreases in NO and NO 2 by up to 1.11 and 4.28 ppb, respectively.In addition, most of the lost NO x was transformed into total nitrate, which increased by as much as 14.92 µg m −3 (Fig. 8e).These results are similar to those of Lowe et al. (2015), who suggested that the nitrate in PM 10 was enhanced by up to 31.4 % (increasing from 3.5 to 4.6 µg kg −1 at night) after considering the heterogeneous uptake processes of N 2 O 5 .The N 2 O 5 uptake increased the O 3 levels by as much as 3.10 ppb in both urban and rural areas in the PRD and decreased the O 3 by up to 1.47 ppb in remote areas (Fig. 8g), which are similar to the findings of Riemer et al. (2003), which indicated that the N 2 O 5 uptake resulted in an increase in the O 3 level in high-NO x areas and a decrease in low-NO x areas.The Cl activation decreased the total nitrate by up to 2.35 µg m −3 (Fig. 8f) because a fraction of N 2 O 5 was consumed to produce ClNO 2 in competition with nitrate production.The simulated O 3 was significantly increased throughout the domain by as much as 4.54 ppb (Fig. 8h), which could be attributed to the release of NO 2 molecules that subsequently undergo photolysis and form O 3 , and the effects of the activation of Cl radicals that initiate VOCs degradation and O 3 formation (Simon et al., 2009).The increase in the O 3 further enhanced the oxidation of NO into NO 2 , which decreased NO by as much as 0.96 ppb (Fig. 8b) and increased NO 2 in urban areas by up to 0.72 ppb (Fig. 8d).The recycling  of NO 2 via ClNO 2 photolysis also contributed to the increase in NO 2 .The O 3 increase also elevated the OH radical levels, which enhanced the loss of NO 2 through its reaction with OH, resulting in the decrease of NO 2 in suburban and rural regions by up to 0.53 ppb (Fig. 8d).

Impacts of N 2 O 5 uptake and Cl activation on NO y partitioning
The composition and partitioning of NO y affect the spatial range that nitrogenous species can reach after emission, and they are therefore of great importance in atmospheric chemistry (Bertram et al., 2013).The average concentration of NO y changed modestly within the PBL over domain 2 with the addition of N 2 O 5 uptake and Cl activation (12.24 and 11.42 ppb in the base and HET+Cl cases, respectively).The fractions of each species in NO y , however, were substantially affected.The NO y partition was calculated over domain 2 for the base and HET+Cl cases (see Fig. 9).The percentage of N 2 O 5 in NO y decreased from 7.80 % in the base case to 1.01 % in the HET+Cl case, and that for NO 3 decreased from 0.38 to 0.09 %.The N 2 O 5 uptake and Cl activation reduced the fraction of NO from 9.59 to 6.84 % and that of NO 2 from 51.07 to 35.17 %.The percentage of total nitrate (nitrate + HNO 3 ) in NO y was significantly increased from 27.5 (i.e.9.6 % + 17.9 %) to 48.6 % (i.e.16.0 % + 32.6 %).The added processes also introduced a new NO y species, ClNO 2 , which accounted for 3.47 % in the HET+Cl case.The decrease in the NO 3 level caused by N 2 O 5 heterogeneous uptake would suppress the night-time chemistry of NO 3 and VOCs.The N 2 O 5 uptake transferred a significant portion of NO x to total nitrate, reducing the lifetime and the reaching range of NO xenriched plumes and thus affecting the NO x -VOCs-O 3 photochemistry.The new species in the HET+Cl case, ClNO 2 , contributed a non-negligible part of NO y and extended the lifetime and reaching range of reactive nitrogen.

Sensitivity of ClNO 2 concentration to chlorine emission
The production of ClNO 2 depends on the chloride concentration in aerosol according to the parameterisation used in this study (Bertram and Thornton, 2009).The only available chlorine EI for China is taken from a global dataset with a relatively low resolution (1 • × 1 • ) and for the year of 1990 (Keene et al., 1999).To test the sensitivity of the ClNO 2 production to Cl emissions, we conducted two simulations in which the RCEI emissions were reduced by half (HET+Cl+0.5RCEI)and doubled (HET+Cl+2.0RCEI).The simulations show that the ambient chloride concentrations responded almost linearly to the applied chlorine emissions (data not shown).The simulated ClNO 2 has a similar temporal pattern in different Cl emissions (Fig. S9).The ClNO 2 concentrations have positive but not linear correlation with Cl emission changes.As shown in Fig. 10, halving the Cl emissions leads to a 31 % reduction in the simulated ClNO 2 level, whereas doubling the Cl emissions results in an average 31 % increase of ClNO 2 .The results indicate that simulation of ClNO 2 production is sensitive to chlorine emission.Therefore, future studies are needed to develop an up-to-date anthropogenic chlorine EI in China to better model ClNO 2 production and to quantify its impact on atmospheric chemistry and air quality.Previous research on chlorine emission could be used as a reference to develop a chlorine EI in China.Chang et al. (2002) compiled a chlorine EI for Houston including the emissions from cooling towers, sea salt, point sources and various water treatments.Sarwar and Bhave (2007) integrated a chlorine EI for the eastern United States with sources of various industrial point sources, area sources, cooling towers and swimming pools.

Summary and conclusions
In this study, a state-of-the-art chemical transport model (WRF-Chem) was further developed by incorporation of a widely used aerosol thermodynamics model (ISORROPIA II), parameterisation of heterogeneous uptake of N 2 O 5 and ClNO 2 production and gas-phase chlorine chemistry.The revised model was used to simulate the spatial distributions of N 2 O 5 and ClNO 2 and the impacts on O 3 and NO y partitioning over the HK-PRD region where high levels of N 2 O 5 and ClNO 2 had been observed.The revised model was able to capture the temporal patterns and the magnitudes of the observed N 2 O 5 and ClNO 2 at a mountaintop site in Hong Kong, but it tended to underpredict N 2 O 5 and overpredict ClNO 2 because of the underestimates of N 2 O 5 sources and overestimates of N 2 O 5 sink and ClNO 2 production.Model simulations show that under average conditions, high values of N 2 O 5 and ClNO 2 are concentrated in the region southwest of the urban areas of the PRD and vertically peak within the layer of 400-800 and 0-200 m a.g.l, respectively.In the night of 3/4 December when the highest ever reported ClNO 2 (4.7 ppb) was observed, the model suggested that the high levels of ClNO 2 were concentrated in the residue layer (∼ 300 m a.g.l.) above the study region.The model simulations suggested that the region downwind of the urban PRD may experience higher levels of ClNO 2 than those observed at the TMS site.N 2 O 5 uptake and Cl activation significantly www.atmos-chem-phys.net/16/14875/2016/Atmos.Chem.Phys., 16, 14875-14890, 2016 decreased the levels of NO and NO 2 by up to 1.93 (∼ 7.4 %) and 4.73 ppb (∼ 16.2 %), respectively, but they increased the total nitrate level by as much as 13.45 µg m −3 (∼ 57.4 %) and the O 3 by up to 7.23 ppb (∼ 16.3 %) within the PBL.Our results demonstrate the significant impacts of N 2 O 5 uptake and ClNO 2 production on NO x lifetime, secondary nitrate production and O 3 formation, and they underscore the necessity of considering these processes in air quality models.Our simulations of ClNO 2 levels over southern China are sensitive to chlorine emissions, which suggests the need to develop a more reliable emission inventory of chlorine for better quantification of N 2 O 5 / ClNO 2 chemistry and its impacts over China.
Both the data and source code of the revised model used in this study are available upon request from the corresponding author (cetwang@polyu.edu.hk).
The Supplement related to this article is available online at doi:10.5194/acp-16-14875-2016-supplement.

Figure 1
Figure 1.(a) Domain settings of WRF-Chem simulations and (b) the terrain in domain 2 (southern China).The red dotted line represents the vertical domain that intercepts the most polluted PRD region along the prevailing wind direction (north-east).TMS is the location of the site where N 2 O 5 and ClNO 2 were measured.HK and PRD are the general locations of Hong Kong and the Pearl River Delta region.

Figure 2 .
Figure 2. Comparison of simulated and observed average (a) N 2 O 5 and (b) ClNO 2 concentrations each night at the TMS site.Error bars represent the standard deviation.

3. 2
Simulation of N 2 O 5 and ClNO 2 with N 2 O 5 uptake and Cl activation 3.2.1 Comparison of simulated N 2 O 5 and ClNO 2 with observation The average observed and simulated (HET+Cl case) concentrations of N 2 O 5 and ClNO 2 were calculated for each night, as shown in Fig. 2. The mean observed N 2 O 5 concentrations for each night varied from 0.02 to 0.74 ppb during the study period, while the average simulated N 2 O 5 values from the HET+Cl case were between 0.02 and 0.35 ppb.The HET+Cl case reproduced the order of N 2 O 5 concentrations but underestimated them within a factor of 3. Differences between the HET case and HET+Cl case in the simulated N 2 O 5 were unnoticeable.For ClNO 2 , the average observed concentrations varied from 0.01 to 0.39 ppb, whilst the mean simulated values for each night varied between 0.05 and 0.42 ppb.The HET+Cl case reproduced the order of ClNO 2 concentrations with an overestimate, mostly within a factor of 4. The simulated and observed hourly concentrations of N 2 O 5 and ClNO 2 are shown in Fig. S3, indicating that the HET and HET+Cl case captured the temporal variations of these two compounds in general.The underpredicted N 2 O 5 and overpredicted ClNO 2 values in the HET+Cl case point to the underestimation of the sources and/or the overestimation of the sink of N 2 O 5 and the overestimation of the production of ClNO 2 .As shown in Sect.3.1, the simulated NO 2 and O 3 levels in the HK-PRD region are lower than the observations, which results

Figure 3 .
Figure 3. Horizontal distributions of (a) N 2 O 5 and (c) ClNO 2 average mixing ratios (ppb) during night-time (18:00-07:00 LT) within the PBL from the HET+Cl case; vertical distributions of (b) N 2 O 5 and (d) ClNO 2 average mixing ratios (ppb) during nighttime (18:00-07:00 LT) in the domain intercepting PRD and along the prevailing wind direction from the HET+Cl case.Red arrows represent the prevailing wind direction.The black shading in vertical plots represents terrain features.

Figure 5 .
Figure 5.The same as in Fig. 4, except in the night of 3/4 December.

Figure 6 .
Figure 6.Horizontal distributions of (a) NO (ppb), (c) NO 2 (ppb), (e) total nitrate (µg m −3 ) and (g) O 3 (ppb) average concentrations during the study period within the PBL from the base case; the average impacts of N 2 O 5 uptake and Cl activation on (b) NO (ppb), (d) NO 2 (ppb), (f) total nitrate (µg m −3 ) and (h) O 3 (ppb) average concentrations during the simulation period in the horizontal domain within the PBL.

Figure 7 .
Figure 7. Vertical distributions of (a) NO (ppb), (c) NO 2 (ppb), (e) total nitrate (µg m −3 ) and (g) O 3 (ppb) average concentrations during the study period in the domain intercepting PRD and along the prevailing wind from the base case; the average impacts of N 2 O 5 uptake and Cl activation on (b) NO (ppb), (d) NO 2 (ppb), (f) total nitrate (µg m −3 ) and (h) O 3 (ppb) average concentrations during the simulation period in the vertical domain.

Figure 8 .
Figure 8.Average impacts of N 2 O 5 heterogeneous uptake on (a) NO (ppb), (c) NO 2 (ppb), (e) total nitrate (µg m −3 ) and (g) O 3 (ppb) average concentrations during the simulation period in the horizontal domain within the PBL; average impacts of Cl activation on (b) NO (ppb), (d) NO 2 (ppb), (f) total nitrate (µg m −3 ) and (h) O 3 (ppb) average concentrations during the simulation period in the horizontal domain within the PBL.

Figure 9 .
Figure 9. Average NO y partitioning during the study period in southern China within the PBL as simulated in (a) the base case and (b) the HET+Cl case.

Table 3 .
Comparison of chemical simulation with observation for the base case and the HET + Cl case.
* Fac2 is defined as the fraction of the simulations that are within a factor of 2 of the observations.

Table 4 .
Comparison of measured and simulated chloride (base case).