Observations of nitryl chloride and modeling its source and effect on ozone in the planetary boundary layer of southern China

Nitryl chloride (ClNO2) plays potentially important roles in atmospheric chemistry, but its abundance and effect are not fully understood due to the small number of ambient observations of ClNO2 to date. In late autumn 2013, ClNO2 was measured with a chemical ionization mass spectrometer (CIMS) at a mountain top (957 m above sea level) in Hong Kong. During 12 nights with continuous CIMS data, elevated mixing ratios of ClNO2 (>400 parts per trillion by volume) or its precursor N2O5 (>1000 pptv) were observed on six nights, with the highest ever reported ClNO2 (4.7 ppbv, 1 min average) and N2O5 (7.7 ppbv, 1 min average) in one case. Backward particle dispersion calculations driven by winds simulated with a mesoscale meteorological model show that the ClNO2/N2O5‐laden air at the high‐elevation site was due to transport of urban/industrial pollution north of the site. The highest ClNO2/N2O5 case was observed in a later period of the night and was characterized with extensively processed air and with the presence of nonoceanic chloride. A chemical box model with detailed chlorine chemistry was used to assess the possible impact of the ClNO2 in the well‐processed regional plume on next day ozone, as the air mass continued to downwind locations. The results show that the ClNO2 could enhance ozone by 5–16% at the ozone peak or 11–41% daytime ozone production in the following day. This study highlights varying importance of the ClNO2 chemistry in polluted environments and the need to consider this process in photochemical models for prediction of ground‐level ozone and haze.


Introduction
Nitryl chloride (ClNO 2 ) is produced from heterogeneous reaction of N 2 O 5 on chloride-containing aerosol [Finlayson-Pitts et al., 1989]. ClNO 2 can potentially impact the oxidative capacity of the atmosphere by production of highly reactive chlorine radical (Cl) and recycling NO x after its photolysis.
The abundance of ClNO 2 in the ambient atmosphere was first measured by Osthoff et al. [2008] on a ship along the Gulf Coast of the U.S., in and around the area of Houston, Texas, and was since detected off the coast of California [Riedel et al., 2012] and in Los Angeles [Mielke et al., 2013]. The presence of ClNO 2 in the polluted marine atmosphere arises from strong emissions of NO x from urban/industrial sources on the coast and of chloride from sea sprays . Measurements at inland locations have later observed elevated ClNO 2 in Boulder, Colorado, (up to 450 parts per trillion by volume (pptv)) [Thornton et al., 2010;Riedel et al., 2013], in Hessen, Germany, (up to 800 pptv) [Phillips et al., 2012], and in London, (up to 724 pptv) [Bannan et al., 2015], whereas moderate levels of ClNO 2 (~250 pptv) were observed during a short field study in Calgary, Canada [Mielke et al., 2011]. Two nonoceanic sources-coal-fired power plants and waste-water treatment facilities-are thought to have provided chloride which led to the formation of ClNO 2 in Boulder and Calgary, whereas ocean sprays are suggested to be the main source of chloride in London and Germany during the periods of elevated ClNO 2 .
Due to photolysis of ClNO 2 and its precursors (N 2 O 5 /NO 3 ) by sunlight, ClNO 2 generally shows very low concentrations during daytime, with elevated concentrations typically observed from sunset through early morning in air containing significant amount of O 3 and NO 2 , as the result of following gas-phase reactions: N 2 O 5 undergoes heterogeneous reaction (R1) to produce ClNO 2 in parallel with hydrolysis of N 2 O 5 : Laboratory and field studies indicate that the product yields of ClNO 2 from the N 2 O 5 heterogeneous reactions are dependent on the relative content of chloride to that of water in aerosol [e.g., Mielke et al., 2013;Roberts et al., 2009], with field observed ClNO 2 yields ranging from 0.07 to 0.98 [Mielke et al., 2011;Osthoff et al., 2008;Riedel et al., 2013;Thornton et al., 2010;Wagner et al., 2013].
The impact of ClNO 2 on radical budget, VOC oxidation, and ozone formation have been assessed using chemical box models which are constrained by measurements of ClNO 2 and other related parameters [Bannan et al., 2015;Osthoff et al., 2008;Riedel et al., 2014;Young et al., 2012;Xue et al., 2015] or emission-driven chemical transport models. The latter first adopted surrogate gas-phase reaction to represent ClNO 2 production [Simon et al., 2009], later the N 2 O 5 uptake coefficient derived from ambient observation and assumed ClNO 2 yield [Simon et al., 2010], and more recently laboratory-derived parameterization of N 2 O 5 uptake and ClNO 2 yield [Sarwar et al., 2012;Sarwar et al., 2014]. Box models have found that chlorine atoms from photolysis of ClNO 2 contribute to daytime net production of ozone of~12 ppbv in Los Angeles [Riedel et al., 2014] and 4% of alkane removal on average (up to 15% for the highest ClNO 2 case) in London [Bannan et al., 2015]. Recent chemical transport model simulations by Sarwar et al. [2014] suggest high concentrations (in the range of parts per billion by volume) of ClNO 2 in several regions of the world and enhancement of daily 8 h ozone of up to 7 ppbv, with the largest concentrations and impact in China and western Europe.
Despite the advances achieved in the previous studies, a full picture of global importance of the ClNO 2 chemistry is still lacking due in part to limited number of observations of atmospheric abundance of ClNO 2 and related chemical and physical parameters in different parts of the world. Up to now, there were only two brief reports of preliminary measurements of ClNO 2 in Asia (Hong Kong) Wang et al., 2014]. In view of large emissions of NO x and aerosol in China's polluted regions, it is of great interest to investigate the heterogeneous reactions of N 2 O 5 and their role in chorine activation and oxidation processes.
This study presents the measurement results of ClNO 2 and related chemical constituents obtained at a mountain top site in Hong Kong during late autumn of 2013. Hong Kong and the adjacent Pearl River Delta (PRD) have long suffered severe pollution from ozone and haze [Chan and Yao, 2008;Wang et al., 2009;Xue et al., 2014a]. Large emissions of NO x , particles, and other anthropogenic pollutants, coupled with the region's proximity to the south China Sea makes it an ideal place to study the heterogeneous formation of ClNO 2 and its impact on ozone (and particulate matter). Indeed, in this field study, we observed the highest concentrations ever reported for ClNO 2 (and N 2 O 5 ) at one night. This is the first observation of ClNO 2 in the upper part of the planetary boundary layer (PBL) over a polluted region of Asia. We first give an overview of the Journal of Geophysical Research: Atmospheres 10.1002/2015JD024556 ClNO 2 results and then examine the chemical characteristics of six high ClNO 2 and N 2 O 5 cases. Highresolution meteorological model coupled with a backward particle dispersion model is used to reveal the atmospheric dynamics which transports plumes to the mountain top. An explicit chemical mechanism is then employed to evaluate the impact of an extensively processed PRD plume at night on the ozone concentrations in the following day. A companion paper [Brown et al., 2016] analyzes the characteristics of NO 3 and N 2 O 5 and their reactions through VOC oxidation and heterogeneous uptake.

Field Study Site
The measurements took place at the peak of Mt. Tai Mo Shan (TMS, 22.410°N, 114.125°E; 957 m above sea level), which is the highest point of Hong Kong. As shown in Figure 1, TMS (and Hong Kong) is in the southeast part of the PRD region and is surrounded by major urban areas in Hong Kong and to the south of Shenzhen and other cities of the Guangdong Province. The measurement site is situated in a natural reserve with restricted access to vehicles. The field campaign was conducted from 15 November to 6 December 2013. The instruments are housed in a shelter on the top of TMS.

Instruments 2.2.1. Chemical Ionization Mass Spectrometry
ClNO 2 and N 2 O 5 were concurrently measured by an iodide-chemical ionization mass spectrometer (CIMS) (THS Instruments Inc., Atlanta), in which the molecules of interest are ionized to iodide clusters followed by their detection with a quadrupole mass spectrometer. The original configuration [Slusher et al., 2004] (with a heated inlet and detection of NO 3 À at 62 amu) was modified prior to the field study because the previous setup was found to suffer from large interference under high NO x conditions . The modified CIMS adopts an unheated inlet and detects ClNO 2 and N 2 O 5 as ion clusters, I(ClNO 2 ) À and I(N 2 O 5 ) À , as recommended by Kercher et al. [2009]. The measurement principle of the CIMS is briefly described here. An alpha radioactive source, 210 Po (NRD, P-2031(NRD, P- -2000, generates the primary ions, iodide ions (I À ), from CH 3 I which is produced by mixing 15 cm 3 min À1 STP of CH 3 I from a cylinder (0.03% vol/vol in N 2 , Arkonic, U.S.) with 1.0 standard liters per minute (sLpm) flow of N 2 (purity: 99.999%). Iodide-water cluster ions are firstly formed ((R13)) then ionize ambient N 2 O 5 and ClNO 2 ((R14) and (R15)), which are detected as I(N 2 O 5 ) À at 235 amu and I(ClNO 2 ) À at 208 amu.
Ambient air was drawn through a 6 m FEP-Teflon tubing (0.25 inch OD) at a total flow of 10 sLpm. The sample inlet was set at 2 m above the shelter roof and is about 6 m above the ground level (see Figure S1 in the supporting information (SI)). The total residence time was less than a second. Only 1.8 sLpm were drawn into the instrument through an orifice, while the rest of the flow was dumped. To reduce uptake of sampled air on particles deposited in the sampling system, a "new" sampling line was used every day. The used tubing and fittings were repeatedly washed with detergent (Decon90), ethanol (Sigma-Aldrich, 99.8%), and ultrapure deionized water under ultrasonic bath (50°C) and then oven dried overnight prior to usage. On-site tests showed that this practice limited the loss of N 2 O 5 on the particles to about 10%.
The potential interferences at 235 amu and 208 amu were determined during measurements for 5 min each hour through bypassing the ambient air into a 25 cm length of 0.5 inch OD stainless steel tubing which was packed with stainless-steel wool and heated at 200°C [Kercher et al., 2009;Phillips et al., 2012]. At sampling flow rate of 1.8 sLpm, the residence time of the ambient air in the zeroing tube was~0.7 s. Laboratory tests show that this zeroing method has removed >98% of N 2 O 5 and ClNO 2 . Figure S2 shows an example of the mass spectrum for ambient samples and for the zeroing period. The sensitivities of the CIMS toward N 2 O 5 were determined by the standard addition method. Specifically, a 2.5 ppbv N 2 O 5 standard was added in the ambient samples at the top of the inlet for 10 min every 3 h to determine and correct for the changes in sensitivity due to variations of ambient conditions, especially changes in humidity. The N 2 O 5 standard was produced by reacting NO 2 in excess with O 3 [Bertram et al., 2009] in a perfluoroalkoxy-made chamber. The concentrations of the produced N 2 O 5 were calculated from decrease in NO 2 concentrations . To check potential loss of N 2 O 5 via productions of nitric acid on surfaces in the calibration system, we measured NO y output with and without a Nylon filter (pore size = 0.02 μm; Pall Corp.) which would absorb HNO 3 gas. The results did not show any detectable difference in the two setups suggesting little production of HNO 3 from N 2 O 5 . The N 2 O 5 content was also determined during the present field study by a Cavity Ring Down Spectrometer (CRDS) [Dubé et al., 2006;Wagner et al., 2011]. The two methods (i.e., calculation from the NO 2 decrease and quantification of the produced N 2 O 5 ) gave similar N 2 O 5 concentrations from the N 2 O 5 source (difference < 3%).
The sensitivities of ClNO 2 were determined every 2 days by dilution of a ClNO 2 standard with humidified dry zero air. ClNO 2 was produced by flowing 150 cm 3 min À1 STP of known concentration of N 2 O 5 through a 0.5 inch OD Teflon tubing (length = 25 cm) quarter filled with slurry of sodium chloride (NaCl). The NaCl slurry was prepared by adding a small amount of ultrapure deionized water (~3 mL) into 1 g of solid NaCl (Sigma-Aldrich, ≥99.8%) producing saturated mixtures. The conversion efficiency of N 2 O 5 to ClNO 2 on the deliquesced NaCl was shown to be unity [Finlayson-Pitts et al., 1989;Roberts et al., 2009]. The loss of the produced ClNO 2 in the system were presumed negligible since its uptake on the water or neutral NaCl slurry is inefficient [Rossi, 2003;Roberts et al., 2008].
The N 2 O 5 and ClNO 2 detections in our configuration show dependence on humidity, similar to other CIMS [Kercher et al., 2009;Mielke et al., 2011]. Results of laboratory tests for our CIMS are given in Figure S3 to show sensitivity as a function of the relative humidity (RH) in the range of 0-90%. It shows that the sensitivity for both N 2 O 5 and ClNO 2 increases with increasing of RH, reaches the highest at 50%-70% of RH and then decreases. The standard addition method for N 2 O 5 calibration has taken account of this humidity dependence at the time of calibration (every 3 h). The changes of sensitivity between the calibrations were corrected based on concurrent measurements of ambient RH and the test result ( Figure S3a). ClNO 2 was calibrated at a fixed RH of 45%, and the sensitivity during the measurement period was corrected from Journal of Geophysical Research: Atmospheres 10.1002/2015JD024556 the humidity measurements and the test result ( Figure S3b). The ambient RH during the measurement period mostly falls in the optimal range.
The detection limit of the instrument was estimated to be 4 pptv for N 2 O 5 and 2 pptv for ClNO 2 (1σ, 1 min averaged data), with a precision of about 10% and an accuracy of ±20%. The CIMS instrument was not at its optimal operating conditions during the field study due to a degraded radioactive source.
Further verification of the ambient measurements of N 2 O 5 by the CIMS was performed by comparing the results with those from the NOAA-CRDS. Figure 2a shows the scatterplot of the data from the two instruments based on the 1 min data collected in the 12 days, and Figure 2b gives the time series of the data for 1-2 December. Overall, an excellent agreement is found, with a correlation coefficient (R 2 ) of 0.93, a slope of 0.99, and an intercept of 30, based on a reduced major axis (RMA) regression. We also made sure that the measured signal at 208 amu was ClNO 2 by examining the correlation with its isotopic ion at 210 amu (I 37 ClNO 2 À ). The slope of the correlation was 0.33 (R 2 = 0.96, see Figure S4), which is approximately equal to the expected theoretical value of chlorine isotopic ratio of 0.32.

Other Measurements 2.2.2.1. Trace Gases
Nitrogen oxides (NO and NO 2 ) were measured with a chemiluminescence instrument (Thermo, Model 42i) equipped with a photolytic converter which is not sensitive to reactive nitrogen compounds other than NO 2 [Xu et al., 2013], and total reactive nitrogen (NO y ) was determined by another chemiluminescence analyzer (Thermo, Model 42i) with an externally attached molybdenum oxide (MoO) catalytic converter. The whole air samples were collected hourly during the daytime on selected days with evacuated electropolished stainless steel canisters. The canisters were later shipped to the University of California, Irvine, laboratory for chemical analysis of methane and nonmethane hydrocarbons [Blake et al., 1994;Simpson et al., 2010]. Nitrous acid (HONO) was measured with a long path absorption photometer instrument (QUMA, Model LOPAP-03) [Xu et al., 2015;Zha et al., 2014]. Ozone (O 3 ) was measured by a UV photometric analyzer (Thermo, Model 49i), while carbon monoxide (CO) and sulfur dioxide (SO 2 ) were measured by Airpointer compact monitoring system (Recordum).

Aerosol Size and Composition
Particle size distribution and numbers were measured in real-time by an Ultrafine Particle Monitor (TSI Model 3031). It measured at a 10 min interval the count of particles in six size bins, i.e., 20-30, 30-50, 50-70, 70-100, 100-200, and 200-1000 nm, with RH ≤ 50% controlled by a nafion dryer. Aerosol surface areas were calculated based on the particle number and geometric mean diameter in each size bin. For the largest size bin of 200-1000 nm, the geometric mean diameter (GMD) of 447 nm may lead to overestimation in surface area. Our previous measurement of the particle size distribution with more size bins at a suburban site in Hong Kong [Guo et al., 2014;Xu et al., 2015] showed that the surface area-weighted mean diameter for 200-1000 nm particles was around 290 nm, and the surface area contributed by particles larger than 400 nm accounted for less than 15% of the total surface area for most of the time. Therefore, a mean diameter of 282 nm (GMD for 200-400 nm) was used to determine the aerosol surface area of 200-1000 nm particles in the present work. In addition, the particle diameters were corrected for particle hygroscopicity by growth factors under high relative humidity condition (RH > 50%). The wet diameters of particles were calculated with kappa-Köhler function, with mean kappa values of 0.3 adopted from ambient measurement at a coastal suburban site in Hong Kong . Filter sample was collected twice a day with a Harvard honeycomb denuder filter-pack system (Chemcomb Model 3500) on selected days (during daytime and at night). The collected filters were sent to laboratory for chemical analysis for water-soluble ionic concentrations using an ion chromatography (DIONEX, ICS3000). Details on sampling and laboratory analysis were described in Nie et al. [2010].

Meteorological Parameters
Ambient temperature, relative humidity, wind speed, and direction were monitored using a relative humidity (RH)/temperature probe and a wind monitor (Gill, UK). The photolysis frequency of NO 2 (J NO2 ) was obtained using a filter radiometer (Meteorologie consult gmbh, Germany).

Meteorological and Chemical Models 2.3.1. WRF
A Weather Research and Forecasting (WRF) model was adopted to simulate three-dimensional meteorological fields and the planetary boundary layer heights of the PRD region and southern China.
The WRF model was run in four two-way nested domains, with a grid resolution of 27 km, 9 km, 3 km, and 1 km, covering East Asia, southern China, Pearl River Delta, and Hong Kong, respectively. Each domain has 31 vertical levels from surface to 100 hPa. The selections of parameterization options of Ahmadov et al.
[2012] were adopted in our study. FNL Operational Global Analysis data set provided by National Center for Environmental Prediction (Available at http://rda.ucar.edu/datasets/ds083.2/) was used to initiate the simulation and to provide the boundary conditions. Nudging-based four-dimensional data assimilation technique in Zhang et al. [2015] and 3-hourly surface and sounding observations obtained from China Meteorological Administration (CMA) were used to improve the simulation results.
Validation of meteorological simulations of the WRF was conducted with hourly surface observation data obtained from CMA and the Hong Kong observatory. The mean bias, correlation coefficient, and rootmean-square error between observations and modeling results were computed for temperature, relative humidity, latitudinal, and longitudinal wind speeds (Table S1 in SI). The results show that WRF model satisfactorily simulated the atmospheric conditions during the campaign period.

HYSPLIT Model
A Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) model was used to analyze the potential source region of an air mass at a receptor backward-in-time. The model can simulate both trajectory and particle dispersions [Draxler et al., 2014]. Trajectory simulation gives the most probable pathway of the air mass and is suitable for conditions under uniform air flow. The particle dispersion application splits air masses into thousands parcels and track their positions (by releasing particles at a receptor and tracking their positions backward in each hour); hence, the particle dispersion is more suitable to simulate air conditions during abrupt synoptic changes, as in our case. In the present study, the HYSPLIT was driven by the WRF generated hourly meteorological fields and was run 12 h backward in time with 2500 particles released at the location of the measurement site. The hourly positions of these particles during the 12 h period were then used to represent the history of air masses. The combination of the high-resolution meteorological simulations and the adoption of particle dispersion are believed to be a better approach in understanding the wind flows and origin of the air masses than use of regular wind measurements or single trajectories with coarse wind data in our study region, which has complex topography (land and oceans, mountains and flatlands, built-up and forested areas, etc.).

MCM Box Model
A comprehensive chemical box model was utilized to evaluate the potential impacts of ClNO 2 on the following day's ozone production. The model was based on the Master Chemical Mechanism version 3.2 Saunders et al., 2003] but was updated to include the heterogeneous reactions of NO 3 and N 2 O 5 and the reactions of chlorine atoms with various primary VOCs, which were not represented in the original Master Chemical Mechanism (MCM) (which only considers reactions of chlorine with alkanes). The updated model contains additional 205 Cl-initiated degradation reactions of all primary alkenes, aromatics, aldehydes, ketones, alcohols, selected organic acids, and nitrates as well as the inorganic chemistry of Cl Journal of Geophysical Research: Atmospheres 10.1002/2015JD024556 precursors. The kinetics data are mostly obtained from the IUPAC database. The more detailed description of the model is presented in a separate work . The partially updated MCM had been used to assess possible impact of ClNO 2 on ground-level ozone with assumed concentrations of ClNO 2 [Xue et al., 2014b]. In the present study, constant value of uptake coefficient was adopted: 0.004 for NO 3 and 0.01 for N 2 O 5 ; the latter is consistent with measurement derived value at the site [Brown et al., 2016].
The model was initialized by the observations of O 3 , NO, NO 2 , HONO, N 2 O 5 , ClNO 2 , CO, SO 2 , CH 4 , C 2 -C 10 nonmethane hydrocarbons, and meteorological parameters. The model was run for a 24 h period to simulate the chemical evolution of the prescribed plume after leaving the TMS site. Two scenarios with and without ClNO 2 were conducted to examine the impact of reactions of Cl atom with VOCs and of cycled NO 2 from photolysis of ClNO 2 . The without ClNO 2 case simulates the scenario with no ClNO 2 produced from the N 2 O 5 heterogeneous uptake (i.e., assuming N 2 O 5 is converted only to HNO 3 which does not participate in daytime photochemistry in the following day). The difference in the model simulated chlorine atom, OH, HO 2 , and ozone concentrations between the two model runs indicate the impact of the ClNO 2 photochemistry. The photolysis frequencies were scaled according to the measurements of J NO2 at the TMS site. The model was run 5 times to stabilize the unmeasured species (e. g., radicals and reaction intermediates), and the daytime output of the last run was subject to further analysis.

Overview of ClNO 2 Abundance
During the field campaign, the CIMS had continuous data on 12 days. The data were mostly in nighttime, as the CIMS instrument was under calibration, testing, or maintenance in most of daytime (except weekends). Figure 3 depicts its time series of ClNO 2 and some chemical and meteorological parameters relevant to its production and loss on 15-17 November and 31 November to 6 December. The data for the other four nights are not shown because their observations can be represented by the cases in Figure 3 (on 23 and 28 November, the site was consistently immersed within cloud, and the ClNO 2 mixing ratios ranged from~50 to 200 pptv, similar to that of 3 December; on the nights of 25 and 29 November the site received air from the east with ClNO 2~5 0 pptv, RH~40-60% and low to moderate levels of ozone and other trace gases, similar to 4 December). Figure 3 indicates that site-intercepted air masses containing elevated levels of ClNO 2 (>400 pptv) or N 2 O 5 (>1000 pptv) in the evenings (between sunset and 23:59, local time) of 15 and 16 November, 31 November, 1 December, early morning (00:00 to sunrise) of 4 December, the evening of 5 December, and early morning of 6 December. In most of these cases (except for 4 December), the ClNO 2 and N 2 O 5 levels started to increase after sunset (see Figure 4 for average profile on these days); elevated mixing ratios of ozone (>80 ppbv) and other air pollutants were observed in preceding daytime, and the photochemical "age" of the plume was relatively young as indicated from moderate NO x to NO y ratios (~0.4). These observations suggest that the ClNO 2 and N 2 O 5 in the early evening were produced from NO x and O 3 emitted/produced in nearby urban areas. The case for 4 December is unique in that (1) the polluted air mass arrived at the site in a much later period of the dark period (3-6 A.M.), (2) 1 min averaged ClNO 2, and N 2 O 5 mixing ratio reached up to 4.7 ppbv and 7.7 ppbv, respectively, in different periods of this plume, and (3) the air mass was much more aged (as indicated by NO x /NO y ratios in Table 1), and the composition showed drastic changes in short periods.   Extremely high mixing ratios of ClNO 2 and its precursor, N 2 O 5 , up to 4.7 ppbv and 7.7 ppbv (1 min averaged), respectively, were seen between 03:30 and 05:00. To our knowledge, such a mixing ratio is the highest value reported in the literature, indicating rapid production of ClNO 2 through N 2 O 5 heterogeneous uptake in this portion of the plume. Different relationships of these two species were observed between 02:00 and 07:00 (which are labeled as I, II, and III in Figure 5). In Plume I, the two rose almost simultaneously, then Plume II saw a sharp rise in ClNO 2 and a slow increase in N 2 O 5 (indicating more efficient heterogeneous uptake of N 2 O 5 ), and an opposite pattern followed in Plume III (i.e., high N 2 O 5 and low ClNO 2 , indicative of less efficient uptake of N 2 O 5 ). The aerosol surface area and the relative humidity were the highest in Plume II (see Figure 5). The air masses between 02:00 and 07:00 were well processed chemically with the NO x /NO y ratio of 0.17, 0.15, and 0.23 in Plume I, II, and III, respectively. Table 1 summarizes the chemical characteristics of the six cases of high ClNO 2 and N 2 O 5 . The average mixing ratios for O 3 in these cases were from 52 to 77 ppbv and 2.0 to 13.4 ppbv for NO 2 . The mean ClNO 2 fraction in NO y at TMS ranged from 0.006 to 0.118, which is comparable to that observed in Los Angeles and Houston (with the fraction up to 0.15-0.20) [Mielke et al., 2013;Osthoff et al., 2008]. We note that 4 December had the much higher levels of ClNO 2 and N 2 O 5 but with comparable NO y and ozone with those for 30 November case. We interpret that the air mass on 4 December had undergone longer periods of processing at night thus accumulated more abundant N 2 O 5 and ClNO 2 . An examination on the composition of the PM 2.5 sample at this night (18:00-06:00) revealed an abnormal level of chloride. The [Cl À ] to [Na + ] mass ratio was significantly larger than that collected during following daytime (2.48 versus 0.91) and was larger than the mean ratio in sea water (1.8) [Seinfeld and Pandis, 2006]. This result indicates the presence of significant amount of nonoceanic chloride in air masses sampled at the night of 4 December. As shown in the following  section, a back trajectory calculation revealed that the aged air mass at the night of 3-4 December came from inland areas of the PRD, suggesting existence of anthropogenic chloride in the PRD region, possibly from coal-fired power plants as indicated by concurrent increase in SO 2 concentrations ( Figure 5). Such inference is supported by previous PM 2.5 aerosol studies conducted in the region, indicative of elevated chloride levels relative to sodium compared to those from ocean sprays [e.g., Tan Figures 6 and 7 show the wind rose at the site and the mean wind field from the WRF simulation during the campaign period. They indicate that the weather of Hong Kong (and the south China coastal regions) in late autumn was strongly influenced by the winter monsoon with winds predominantly from the east and northeast.

The Origin of the High ClNO 2 and N 2 O 5 Air Masses
To understand the origin of the six high ClNO 2 and N 2 O 5 cases, we have examined the simulated winds and the 12 h backward-in-time particle dispersion results from WRF and HYSPLIT models (see section 2). The results show that the plumes arriving at the site were associated with changing air flow from the prevailing E/NE to a more northerly direction. Figure 8 shows the result for the 4 December case (other cases are given in Figure S5). At midnight (i.e., before the arrival of the ClNO 2 and N 2 O 5 plume), the air traveled from the east; but at 04:00 (the middle of polluted air), one branch of air came from the north and from low altitudes, which would bring pollutants from urban areas in Shenzhen and inland areas of the PRD. The enhanced concentrations of SO 2 , increased relative humidity, as well as elevated NO y (see Figure 5) at 04:00 are consistent with meteorological indication of surface air masses. Other cases show similar (but varying) impact of pollution from the north, and the plumes were transported to the site at earlier hours of the night, which means that the air mass had experienced less nighttime processing, consistent with the chemical indicator of age, the NO x /NO y ratios (see Table 1). Therefore, based on the above chemical and meteorological analysis, we can conclude that most of the cases of elevated ClNO 2 and N 2 O 5 are in an early stage of nighttime processing of daytime photochemical pollutants and the  4 December case (with the highest ClNO 2 and N 2 O 5 ) represents air masses from the PRD region with prolonged nighttime processing. The site intercepted this aged plume as a result of peculiar atmospheric dynamics on that night (i.e., changing wind at a later part of the night). An analysis of a nighttime chemical clock for 4 December based on odd oxygen and discussed in the companion paper [Brown et al., 2016] also suggests a long overnight processing time for plumes sampled during the early hours of 4 December.

Impact of ClNO 2 on Ozone Formation
One important impact of ClNO 2 in extensively processed nighttime plumes is its role in photochemistry and ozone production on the following day during air mass transport to downwind locations. The 4 December case provides an opportunity for such an analysis, as the air mass at the night had undergone prolonged processing (NO x /NO y = 0.18-0.23), and it was measured just a few hours before sunrise. The model was initialized with the ClNO 2 concentrations observed 1-3 h before sunrise (see below) assuming no additional ClNO 2 production prior to sunrise. This additional production of ClNO 2 is difficult to estimate, and we analyze here in terms of the lower limit to the effect on ozone.
To examine whether the ClNO 2 -rich air was subject to dilution at the sampling height due to growth of the PBL in following daytime, we examined the WRF-calculated height of PBL at five locations over downwind maritime areas (see Figures S6 and S7). It can be seen that the sample location was well above the WRF-calculated height of the PBL at five locations between sunrise (200 m) and noon (600 m). Therefore, ClNO 2 -driven photochemistry in the sampled air mass residing within the residual layer is not subject to dilution induced by boundary layer growth before noontime. This residual layer O 3 will presumably mix to surface level as the boundary layer grows during afternoon, but our box model simulations are not designed to capture this effect.
We initiated the box model with the chemical data collected in the two periods (1) 3:30-4:00, i.e., with highest abundance of ClNO 2 (4.7 ppbv) and moderate level of N 2 O 5 (2.15 ppbv), representing a high ClNO 2 -yield case and (2) 5:30-6:00, a period with very high N 2 O 5 (4.96 ppbv) and moderate ClNO 2 (1.10 ppbv) representing a low yield case. Table S3 shows the mixing ratios/values of the chemical and meteorological parameters in the simulations of these two cases. As the canister samples for hydrocarbons were collected only during daytime (due to restriction to access the site at night), we estimate the initial hydrocarbons mixing ratios in the simulated night periods according to the following approach. The NO y mixing ratio at 04:00 was comparable (~40 ppb) to that at 11:00 when a canister sample was collected (see Table S4 for composition). Considering more chemically processed air at 04:00 than at 11:00, as indicated by the NO x /NO y ratios of 0.21 versus 0.57, we assume similar abundance of longer-lived alkanes, alkynes, and benzene but only half of the more reactive alkenes and other aromatics in the early morning. We also conducted simulations assuming same, three fourths and one fourth of the reactive hydrocarbon abundance. The oxygenated organics such as aldehydes and ketones were not measured at TMS but were calculated from the measured hydrocarbons by running the MCM model for 5 days [Xue et al., 2014b]. The high ClNO 2 case has lower NO 2 (5.0 ppbv) and N 2 O 5 levels (2.14 ppbv) compared to the low ClNO 2 case (with 9.62 ppbv of NO 2 and 4.96 ppbv of N 2 O 5 ), which means that the chemistry in the former modeling case has lower a NO x /VOC ratio (because the hydrocarbon abundance is the same in the two cases.) Figures 9 and 10 show the simulated concentrations of ClNO 2 , Cl atom, OH, HO 2 , and ozone as function of time since the plume left the site for the low and high yield case. Including ClNO 2 , photolysis resulted in peak concentrations of Cl of 2.4 × 10 4 molecules cm À3 for the low ClNO 2 yield case and of 1.1 × 10 5 molecules cm À3 for the high ClNO 2 case in the morning period. The Cl reactions with VOCs and the recycled NO 2 have increased the concentrations of OH (via (R7)) and HO 2 (via (R6)) by a maximum of 40% and 53% for the low ClNO 2 case and by 77% and 106% for high ClNO 2 case, respectively. For the latter case, the integrated OH produced from O 3 photolysis during daytime (6:00-18:00 local time) is 2.0 ppbv; production OH from photolysis of HONO is 3.0 ppbv, with OH produced from HO 2 + NO being 18.5 ppbv. Therefore, in the high ClNO 2 case, photolysis of 4.7 ppbv of ClNO 2 has contributed nearly 22% of the total radical production during daytime, indicating significant impact on the oxidizing capacity in that air mass.
The effect on peak ozone in low and high ClNO 2 yield case is also calculated. For the low ClNO 2 case, the peak mixing ratio increased by about 5% (from 99 to 104 ppbv, Figure 9) with lower percentage increases in higher VOC concentrations (see Figure S8). For the high ClNO 2 case, daytime peak ozone increased from 95 ppbv to 110 ppbv, or a 15.8% increase ( Figure 10). The ozone formation in the high ClNO 2 case is not sensitive to different concentrations of VOCs used in the simulations, as it is in a NO x -limited regime indicated by higher ozone levels with increased NO 2 concentrations. The ClNO 2 chemistry accounts for 11% of the total ozone produced (~45 ppbv; defined as the difference between minimum and maxima in Figure 8d) for the low ClNO 2 case but 41% for high ClNO 2 case (~36 ppbv, see Figure 9d). For comparison, previous model calculations showed enhancement of 9-12 ppbv in daytime O 3 peak off the coast of Houston  and in the Los Angeles outflow [Riedel et al., 2014], but these locations had a lower percentage increase in daytime production of ozone compared to TMS due to much higher nighttime ozone levels (60-70 ppbv) at TMS than those (20-30 ppbv) at the previous ground sites. These results suggest variable and potentially significant role of ClNO 2 in the formation of ozone in polluted regions. More studies in lacking regions/periods are warranted in order to get a more complete picture of the effect of ClNO 2 on the formation of secondary pollutants in different chemical environments.

Conclusion
The first measurements of ClNO 2 in the upper planetary boundary layer of polluted regions of China were conducted at the peak of Mt. Tai Mo Shan (957 m asl) in Hong Kong. The results reveal frequent impact of ClNO 2 and N 2 O 5 plumes on the high-elevation site in early periods of night with peak ClNO 2 mixing ratios of 200-1500 pptv (1 min average). The highest ever reported ClNO 2 (4.7 ppbv) was observed in the early morning of one night in extensively processed air from major industrial/urban areas of the Pearl River Delta. Model calculations demonstrate significant enhancement of ozone due to the high levels of ClNO 2 . We believe that similar concentrations and more frequent occurrences may occur in other locations downwind of the PRD pollution sources and in other regions with strong emissions of NO x -aerosol-Cl, such as the North China Plain (Beijing) and the Yangtze River Delta region (Shanghai). We suggest field measurements and modeling studies be conducted in these regions in order to obtain a full picture of the role of N 2 O 5 heterogeneous reactions and chlorine activation in the formation of ground-level ozone and haze in China.