The measurements were conducted at the Tung Chung air quality monitoring station (TC; 113.93∘ E, 22.30∘ N). It is located about 3 km south of the Hong Kong International airport, and is in a residential area of a new town in western Hong Kong (see Fig. S1 in the Supplement). This station is characterized as a polluted receptor site as it receives urban plumes from Hong Kong under easterly winds and regional air masses from the PRD region when northerly winds prevail, and is the location where the maximum O3 levels are usually recorded in Hong Kong (Xue et al., 2014b). Details of this station and analyses of HONO and aerosol data have been described in our previous publications (Xu et al., 2015; Xue et al., 2014b; Zhou et al., 2014).

A 1-month campaign was carried out from 6 August to 7 September 2011, which covered two distinct types of meteorological conditions and air quality (see Fig. 1). For the majority of the campaign, Hong Kong was influenced by clean marine air masses and featured by good air quality (typical summer conditions as a result of the Asian monsoon). In contrast from 25–31 August, a heavy multi-day photochemical smog event hit Hong Kong with northerly winds prevailing during the daytime and elevated concentrations of various air pollutants were observed. In the present study, this episode was subject to a detailed modeling analysis to understand the atmospheric oxidative capacity and ROx chemistry, made possible with the most comprehensive suite of measurements taken for the first time in Hong Kong.

A full suite of trace gases and meteorological parameters were simultaneously measured during this episode (as summarized in Supplement Table S1). Here a brief description is given of the measurements used in the present study. Major air quality target pollutants were routinely monitored with commercial analyzers: O3 with a UV photometric analyzer (TEI model 49i); CO with a non-dispersive infrared equipment (API model 300EU); NO and NO2 with a chemiluminescence analyzer (TEI model 42i) equipped with a selective blue light converter (Xu et al., 2013). NOy was measured by another chemiluminescence instrument (TEI model 42cy) with an external molybdenum oxide (MoO) catalytic converter (Xue et al., 2011). HONO was measured in real-time by a long path absorption photometer (QUMA model LOPAP-03) (Xu et al., 2015). Nitryl chloride (ClNO2) was detected using a custom-built chemical ionization mass spectrometer (CIMS; THS Instruments Inc., Atlanta) (Tham et al., 2014). Peroxyacetyl nitrate (PAN) was measured by the same CIMS instrument with a heated inlet, and the potential interference caused by high NO was corrected based on laboratory tests (Slusher et al., 2004; Wang et al., 2014). Hydrogen peroxide (H2O2) and organic peroxides were measured by an enzyme-catalyzed fluorescence instrument (Aerolaser AL-2021) (Guo et al., 2014). Particle number and size distributions in the range of 5 nm to 10 µm, which were used to calculate the aerosol surface density, were measured with a wide-range particle spectrometer (WPS; MSP model 1000XP) (Gao et al., 2009).

C2–C10 non-methane hydrocarbons were measured at a time interval of 30 min by a commercial analyzer that combines gas chromatography (GC) with photoionization detection (PID) and flame-ionization detection (FID) (Syntech Spectras, model GC955 Series 600/800 POCP). The detection limits for the measured VOCs ranged from 0.001 to 0.19  ppbv. In addition, 24 h whole air canister samples were collected on selected days (e.g. 25 and 29 August) for the detection of C1-C10 hydrocarbons by using GC with FID, electron capture detection (ECD) and mass spectrometry detection (MSD). The analyses were carried out at the laboratory of the University of California at Irvine, and the detection limit was 3 pptv for all measured species (Simpson et al., 2010; Xue et al., 2013). As evaluated in our previous study, both sets of hydrocarbon measurements agree very well apart from the alkenes. Here the real-time data tended to systematically overestimate the canister measurements (Xue et al., 2014b). Considering the generally lower detection limit of the canister observations, the high-resolution real-time data were corrected in the present study according to the canister data. C1-C8 carbonyls were measured by collecting air samples on DNPH-coated sorbent cartridges followed by high pressure liquid chromatography analysis (Xue et al., 2014c). For the carbonyls, a 24 h integrated sample was collected on 25 August, and eight 3 h samples were taken throughout the day on 31 August. The measured hydrocarbon and carbonyl species are listed in Table 1.

Meteorological parameters were monitored by a series of commercial sensors, including a probe for ambient temperature and relative humidity (Young RH/T probe) and an ultrasonic sensor for wind speed and direction (Gill WindSonic). Photolysis frequency of NO2 (JNO2) was measured with a filter radiometer (Meteorologie Consult gmbh). All of the above techniques have been validated and applied in many previous studies, with detailed descriptions of the measurement principles, quality assurance and control procedures provided elsewhere (Guo et al., 2014; Xu et al., 2015; Xue et al., 2011, 2014a, b and c). See also Table S1 for a summary of the measurement techniques/instruments and time resolutions.

The zero-dimensional chemical box model OBM-AOCP (Observation-Based Model for investigating the Atmospheric Oxidative Capacity and Photochemistry) has been utilized in many previous studies to evaluate O3 production (Xue et al., 2013, 2014a, b), PAN formation (Xue et al., 2014c), and oxidative capacity (Xue et al., 2015). Briefly, the model is built on the Master Chemical Mechanism (MCM; v3.2), a nearly explicit gas phase mechanism describing the degradation of 143 primary VOCs (Jenkin et al., 2003; Saunders et al., 2003), and is updated to include both a heterogeneous chemistry scheme (including heterogeneous processes of NO2, NO3, N2O5, HO2, and ClONO2; Xue et al., 2014a) and a chlorine chemistry module that describes the reactions of Cl radical with various VOC compounds (Xue et al., 2015; note that the basic MCM only considers the reactions of Cl radical with alkanes). In addition to the chemistry, dry deposition and dilution mixing within the boundary layer are also included in the model (Xue et al., 2014a). The mixing layer height affecting the deposition rate and dilution mixing was assumed to vary from 300 m at night to 1500 m in the afternoon. Sensitivity model runs with different maximum mixing heights (1000 and 2000 m) indicated that its impacts on the modeling results (e.g. simulated HOx concentrations and OH production rate) were negligible. A detailed description of the model set up is provided in the Supplement.

The model is capable of simulating the concentrations of highly reactive species (e.g. radicals) and quantitatively evaluating several key aspects of atmospheric photochemistry such as oxidant formation (e.g. O3 and PAN), VOC oxidation and radical budgets. In our model, the rates of over 15600 reactions out of the full MCM (v3.2) are individually and instantaneously computed and grouped into a relatively small number of major routes. The calculation of ozone and PAN production rates have been described elsewhere (Xue et al., 2014a, c). Here the emphasis is placed on the computation of AOC and ROx budget. AOC is calculated as the sum of oxidation rates of CO and VOCs by the principal oxidants, namely OH, O3, NO3 and Cl (Xue et al., 2015). The partitioning of the AOC among individual oxidants or VOC groups can be also assessed. The chemical budgets of OH, HO2, and RO2 are quantified by grouping a huge number of relevant reactions into dozens of major production, cycling and loss routes. The principal radical sources in the polluted atmosphere generally include photolysis of O3, HONO, H2O2 and OVOCs as well as reactions of O3+VOCs, NO3+VOCs and Cl+VOCs. The radical sinks mainly include the ROx-NOx and ROx-ROx cross reactions. Besides, a number of other minor reaction pathways were also computed to facilitate a thorough investigation of the ROx chemistry (see Figs. 5 and 7).

The measurement data of O3, NO, NO2, HONO, ClNO2, H2O2, PAN, CO, C1-C10 HCs, C1-C8 carbonyls, aerosol surface area and radius, temperature, RH and JNO2 were averaged or interpolated to a time resolution of 10 min for the model constraints. For carbonyls, the diurnal profiles measured on 31 August 2011, throughout which eight 3 h samples collected were adopted and scaled to the 24 h average data observed on 25 August (see Fig. S4 for the measured profiles of selected carbonyls). An initial concentration of 0.5 ppm of H2 was assumed in the model. Photolysis frequencies were calculated as a function of solar zenith angle within the model (Saunders et al., 2003) and further scaled with the measured JNO2 values. The model starts from 00:00 local time (LT) and runs for a 24 h period. Prior to the formal calculation, the model was run for 5 days with constraints of the campaign-average data to reach steady states for the unconstrained compounds (e.g. radicals). The final outputs were extracted and subject to further analyses.

The measured concentrations of major pollutants and meteorological parameters at TC are depicted in Fig. 1. During 25–31 August 2011, Hong Kong was hit by a prolonged photochemical smog episode, with concentrations of various air pollutants exceeding the ambient air quality standard (the zoomed-in figure of this episode is given in the supplement). Peak O3 mixing ratios of over 150 ppbv were observed almost every day within the 1-week period, except for 29 August when the peak was 135 ppbv. As another indicator of photochemical smog, the concentrations of PAN were also very high with the peak values exceeding 4 ppbv every day (except for 3.7 ppbv on 29 August). The maximum hourly values of O3 and PAN were recorded at 162 and 6.95 ppbv, respectively. Extremely high levels of NOx (peak of ∼ 150 ppbv), CO (peak of ∼ 1000 ppbv) and particulate matter (as indicative of > 500 Mm-1 of aerosol scattering coefficient) were also determined. Overall, inspection of observational data reveals the markedly poor air quality and serious photochemical pollution over the region during the episode.

Table 1 lists the 24 h average concentrations of hydrocarbons and carbonyls measured on 25 and 31 August 2011. It is clearly seen that the VOC levels, in particular for reactive aromatics and aldehydes, were also very high during the episode. On 25 August, for instance, the 24 h average values of toluene, summed xylenes, formaldehyde and acetaldehyde were as high as 9.47, 3.87, 9.89, and 4.25 ppbv, which were 3–30 folders higher than those measured during the non-episode period of the campaign (figures not shown). HONO and ClNO2, two precursors of OH and Cl radicals, were also measured. Elevated HONO (up to 2–3 ppbv) and moderate ClNO2 (up to 0.5–1 ppbv) were usually found at night, and what is more interesting is that the daytime HONO levels were also significant (over 1 ppbv in general; see Fig. 1). Such daytime HONO levels cannot be explained by the known gas-phase source and indicates the existence of other unknown source(s) (Xu et al., 2015), yet exploring the unknown HONO sources is beyond the scope of the present study. High abundances of O3, HONO and carbonyls would definitely lead to strong production of ROx radicals, and the abundant VOCs would facilitate efficient radical propagation (e.g. OH→RO2). Therefore, strong atmospheric oxidative capacity and intensive in situ photochemistry can be expected from the above analyses.

The dynamic cause of this episode was a distant tropical cyclone that introduced warm stagnant weather and facilitated accumulation of air pollutants in Hong Kong and the PRD region. The weather condition featured high temperatures (30–35 ∘C) and relatively low RH (40–80 %; see Fig. S2). During the daytime, the prevailing surface winds were consistently from the northwest with relatively low wind speeds (∼ 2 m s-1), suggesting the transport of processed air masses from the upwind PRD region to the site. This was further confirmed by the 48 h backward trajectories calculated by the HYSPLIT model (Draxler and Rolph, 2016), which indicated that for most days the air masses had spent a large portion of time over the PRD region prior to arriving at TC (Fig. S3).

There was an exception on 25 August when the air flow was switching from southerly maritime air to northerly PRD regional air masses (see Fig. S3). This case is believed to be more influenced by the local air in Hong Kong, because (1) northerly winds during the daytime were somewhat weak compared to the other cases (see Fig. S2); (2) the backward trajectories also indicated less impact from the PRD region (Fig. S3); and (3) the CO / NOy ratio on that day was significantly lower than those on the following days (Fig. S5), which is consistent with the previous finding that the PRD air masses have higher CO / NOy ratios than those from Hong Kong (Wang et al., 2003). The evolution of the CO / NOy ratio clearly indicates the transition from local (25 August) to regional air masses (27–31 August) throughout the 1-week episode (Fig. S5). In the following discussion, detailed modeling analyses are conducted for the 25 and 31 August cases, which are representative of local Hong Kong and regional PRD pollution, respectively.

The strong oxidative capacity of the atmosphere during the pollution episodes was confirmed by quantifying the loss rates of CO and VOCs via reactions with OH, O3, NO3 and Cl, as shown in Fig. 2. The calculated AOC was up to 2.04 × 108 and 1.27 × 108 molecules cm-3 s-1, with daytime averages (06:00–18:00 LT) of 7.26 × 107 and 6.30 × 107 molecules cm-3 s-1, on 25 and 31 August, respectively. As such, the total number of CO and VOC molecules depleted throughout the daytime was 3.14 × 1012 and 2.72 × 1012 per cm-3 of air for both cases. Such levels of AOC at TC are much higher than those determined from a rural site in Germany (Geyer et al., 2001), but a bit lower than that assessed from a polluted area in Santiago, Chile (Elshorbany et al., 2009).

OH was, as expected, the predominant oxidant accounting for 89 and 93 % of the AOC on 25 and 31 August, respectively. NO3 was the second important oxidant with contributions of 7 and 3 % for both cases. In particular, NO3 contributed to 43 % of the AOC at 15:00 LT on 25 August under a weak solar radiation condition. The major fuels for NO3 oxidation were OVOCs (i.e. 77–90 %) and alkenes (10–23 %). In comparison, O3 and Cl (produced from ClNO2 photolysis) had minor contributions due to the relatively lower abundances of alkenes and Cl radicals (i.e. the modelled peak value of Cl was ∼ 1 × 104 atoms cm-3). Overall, the OH-dominated AOC at TC is in line with the previous studies at other urban locales (Elshorbany et al., 2009; Bannan et al., 2015), and the present analysis suggests that the NO3 radical may play an important role in the daytime oxidation under certain conditions (see a detailed evaluation in Sect. 3.4).

We further assessed the loss rates of major VOC groups due to OH oxidation, from which the partitioning of OH reactivity among different VOCs can be elucidated. The results are presented in Fig. 3. OVOCs clearly dominate the OH reactivity with daytime average contributions of 60 and 75 % and with maximums in the afternoon of over 80 % for both cases. Aromatics are the second largest contributor comprising on average 22 and 10 % of the daytime OH reactivity. For the Hong Kong local case on 25 August, especially, aromatics made up the majority (i.e. 40–60 %) of the OH reactivity in the early morning period when there were much fresher air masses. In comparison, alkenes and alkanes only accounted for a small fraction (8–10 %) of the OH reactivity at TC. These results are in fair agreement with the previous studies of Lou et al. (2010) and Whalley et al. (2016), which indicated the dominance of secondary OVOCs in the observed OH reactivity in the PRD region and central London.

As shown above, the partitioning of principal oxidants and OH reactivity is quite similar for both cases. In comparison with the regional case on 31 August 2011, nonetheless, the Hong Kong local case (i.e. 25 August 2011) showed higher AOC levels and more contribution from aromatic VOCs to the OH reactivity. Such difference should be due to the fresher air masses and hence more reactive VOC species during the local case. In the following section, a detailed budget analysis of the radical initiation, recycling and termination processes is presented.

The NO3 radical can initiate the oxidation of VOCs and lead to formation of RO2 and nitrogen-containing organic aerosols (Rollins et al., 2012; Saunders et al., 2003). These processes are usually considered to mainly occur at night and be negligible during the daytime due to the fast photolysis of NO3. In the present study, we observed an interesting case that provided evidence of the operation of daytime NO3-initiated chemistry. The detailed measurement data of chemical and meteorological parameters in this case (i.e. 25 August 2011) are depicted in Fig. 8. During this episode, the air was characterized by high concentrations of O3 (up to 170 ppbv), NO2 (∼ 25 ppbv as the afternoon average) and VOCs (see Table 1). Meanwhile, the solar irradiation arriving at the surface was weaker than other days, as evidenced by the relatively lower values of JNO2 (with a peak of 6.0 × 10-3 s-1) compared to clear days with ∼ 10 × 10-3 s-1 (see Fig. 1). The ambient relative humidity (RH) in the afternoon was in the range of 60–70 %, implying that there was little cloud on the site, whilst the aerosol scattering coefficient was very high (up to 525 Mm-1; compared to 28 ± 12 Mm-1 on clear days). Hence, the attenuated solar radiation is possibly attributed to the abundant aerosol loadings. Under such conditions, the model produced an afternoon peak of NO3 of ∼ 7 pptv at 13:30–15:00 LT (except for the maximum of 11.3 pptv at 14:50 LT that was coincident with an extremely low solar radiation condition).

To further understand the causes and impacts of the daytime NO3 chemistry, a detailed budget analysis was conducted with the OBM-AOCP model. The midday average (09:00–15:00 LT) production and destruction rates of NO3 from the individual reaction pathways are documented in Fig. 9. The co-existence of high concentrations of O3 and NO2 resulted in a very strong NO3 production with an average strength of 11.0 ppb h-1. Given its high reactivity, NO3 once formed, can be readily photolysed as well as react with NO and VOCs. For this case, about 80 % (i.e. 8.8 ppb h-1) of NO3 reacted with NO to convert back to NO2. Due to the weak solar radiation, photolysis only accounted for 6.2 % (or 0.7 ppb h-1) of the NO3 loss. In comparison, reactions of NO3 with VOCs contributed 11.7 % (or 1.3 ppb h-1) to the total loss at midday. During this episode, therefore, NO3 appeared to be the second-most important oxidant (see Sect. 3.2) and the reactions of NO3 with VOCs presented a considerable RO2 source during the daytime (Sect. 3.3). In addition, the NO3-initiated degradation of VOCs could also lead to formation of secondary organic nitrate aerosols, but was not simulated in the present study.

The above analysis indicates the possible importance of NO3-initiated oxidation in the daytime atmospheric photochemistry under specific conditions. This analysis is solely derived from an observation-based modeling study of a unique pollution case in Hong Kong. Nevertheless, we hypothesize that it may also take place in other polluted urban atmospheres, especially in the large cities of China. It is known that eastern China now suffers from widespread and severe photochemical smog during the summer, which features elevated concentrations of O3, NOx, VOCs, and fine particulate matter (Xue et al., 2014a). The intense air pollution usually induces “smoldering” weather with poor visibility and hence attenuated solar irradiation (Ding et al., 2013). All these unfavourable conditions would facilitate the operation of daytime NO3 chemistry as found in Hong Kong in the present study. Further studies are required to verify this phenomenon in other polluted environments and quantify its contributions to the formation of ozone and secondary organic aerosols.