Generally, a city-wide high-PM episode is of more public concern than pollution at a single station. To examine PM pollution throughout the city, we evaluated the hourly PM2.5 data across the 15 HKEPD general air quality monitoring stations. The HKUST observation period spans from 30 August to 31 December 2020; thus, we examined air quality data at the 15 HKEPD stations from 10 July to 31 December 2020. The upper-level wind direction as well as the sea level pressure and dew point date were used to determine the dates by which the seasons were divided (Yu, 2002); the details are given in the Supplement (Text S3). Specifically, the study period spanned three seasons: summer (10 July—7 October), fall (8 October–28 November), and winter (29 November–31 December). Generally, the PM2.5 concentration varied synchronously among different sites, regardless of whether they were urban or background sites, with Pearson correlation coefficients (Rp) ranging from 0.71 to 0.92 during the non-episode period, suggesting the regional characteristics of PM pollution in HK (Fig. S4).

In this work, we define the occurrence of a PM2.5 pollution episode as a PM2.5 concentration higher than 35 µg m-3 for at least 6 consecutive hours at three or more stations. This value (35 µg m-3) is the current annual PM2.5 air quality objective set by the Hong Kong government, which aligns with the World Health Organization (WHO) Interim Target 1 annual PM2.5 concentration value (WHO, 2021). Using this screening criterion, 11 PM episodes were identified in this study: one in summer (EP1), five in fall (EP2–EP6), and five in winter (EP7–EP11). Table 1 lists the statistical summary of the PM2.5 averages during episode and non-episode periods, the meteorological conditions, and the gas pollutants (O3 and NOx) during the examined period (10 July–31 December 2020). Among the identified episodes, five were short episodes, lasting less than 1 d (three of which primarily occurred at night-time – denoted as EP6N, EP7N, and EP11N), and the rest lasted for much longer, ranging from 1 d to 1 week. The highest PM2.5 pollution was observed during EP1 in summer, and the city-wide average concentration was 37.6 µg m-3 during this episode. The lowest PM2.5 pollution occurred during EP9 in winter (28.4 µg m-3). The PM level during episodes was more than 2 times higher than the non-episode average (12.5 µg m-3). Wind speeds were generally higher than 2 m s-1 during the PM episodes, except for during EP1 and EP6N. A high concentration of O3 was observed during EP1–EP6 and EP10, indicating a higher atmospheric oxidation capacity. EP11N showed high concentrations of NOx, which may be attributed to enhanced local vehicle emissions.

Figure 1 shows the spatial variation in the average PM2.5 concentration during individual episodes as well as during the remaining non-episode hours. The maximum-to-minimum (max-to-min) ratio in Table 1 shows the difference in the PM level across the 16 stations by normalizing the maximum average PM concentration against the minimum PM average among the 16 stations for each episode. A ratio close to 1 suggests the uniformity of PM pollution across the whole city, whereas a larger ratio indicates a spatial gradient in PM pollution. A ratio of 1.6 was observed for non-episode periods, with the following spatial pattern being observed: north-western stations (e.g. TM, YL, and TW) > central stations (e.g. KT, TP, and NH) > eastern/south-eastern stations (e.g. TK and SN) (Fig. 1). This pattern suggests the consistent influence of regional transport from the northern PRD region to HK. Episodes during summer to early fall (i.e. EP1–EP4) showed a slightly lower max-to-min ratio (∼1.4), suggesting city-wide pollution characteristics, whereas an enhanced ratio was observed for most winter episodes (1.9–4.3), signifying a larger concentration gradient in winter. The spatial variation during the winter episodes generally followed the trend seen in non-episode periods, with north-western stations showing higher PM levels than the eastern/south-eastern stations, with only a few exceptions being noted. For example, EP7N showed the highest PM level at three northern stations (TM, MB, and TP) and a concurrent high wind speed (6.13 m s-1), signalling regional input from northern inland regions. EP10 showed the largest spatial difference (i.e. a max-to-min ratio of 4.3). During EP10, an enhanced PM level was observed at the three north-western stations (TM, YL, and TC), while PM2.5 at other stations did not show an obvious increase. Such a spatial pattern suggests the influence of local pollution around this district. In contrast, during EP11N, higher concentrations were recorded at stations in the urban centre of HK (i.e. CW, EN, KC, SP, KT, and TK) than at the north-western stations, which may be attributed to enhanced local vehicular emissions during the end-of-year holidays (i.e. 29–30 December). The above results suggest that PM episodes showed more homogeneity during summer to early fall, whereas spatial heterogeneity of the episodes in winter was more notable. The latter implies that air quality monitoring at a single station cannot represent the pollution status of the entire city.

Compared with the total PM mass, the pollution characteristics of individual components, especially organics, may be quite different at different locations. The full PM composition data are only available at the HKUST supersite for the period from 30 August to 31 December 2020. Figure 2 shows the time series of PM2.5 and its major chemical components as well as select individual POA and SOA tracers and their VOC precursors. The source-specific organic markers include (1) BB-derived POA tracer (levoglucosan) and SOA tracer (4-nitrocatechol); (2) anthropogenic SOA tracers, including naphthalene/methylnaphthalene SOA tracer (phthalic acid) and monoaromatic SOA tracer; and (3) biogenic SOA tracers, including six isoprene SOA tracers, six α-pinene SOA tracers, and one β-caryophyllene SOA tracer. The SOA tracers generally showed higher concentrations during the daytime (Fig. S9), which is consistent with their secondary origins. A moderate to good correlation was observed among the SOA tracers and with sulfate, suggesting that this was a regional feature (Fig. S10).

To characterize chemical features in the formation of PM episodes, we examined the PM composition before and during each episode. The campaign-wide average is not representative of the normal condition specific to individual seasons, as the emission sources and meteorological conditions varied among seasons. The selection of the pre-episode period can minimize the interference from different meteorological conditions, such as temperature and boundary layer height, among seasons. This comparison (i.e. the pre-episode period vs. the episode period) can better examine the rapid formation of high-PM episodes. The selection of the pre-episode period for each episode is shown in Table S2 and Fig. S11. We select the pre-episode time windows primarily based on the principle that the pre-episode duration was comparable with that of the episode and that a given time interval (>12 h) exists between the pre-episode period and the episode immediately preceding it (in order to avoid residual influences from the previous episode). The air mass origins for each episode–pre-episode time window pair are shown in Fig. S12. Similar air mass origins, from northern continental regions, were observed during the pre-episode and episode periods for most episodes (except for EP1), excluding a sudden change in the air mass origins as a leading cause of the rapid increase in PM levels. The mass increment ratio (MIR), calculated as the mass concentration during the episode divided by that before the episode, was used to evaluate the change in the concentration of different chemical species during the episode. A MIR >1 suggests an increase in the concentration during the episode and vice versa.

Figure S13 shows the average PM composition for the pre-episode and episode periods. EP6N and EP10 are not obvious at this site, with an average PM <20 µg m-3, which is only slightly higher than the corresponding pre-episode period. The PM composition showed that sulfate and OM were the major components throughout the measurement period, while increased nitrate and ammonium concentrations were recorded in winter due to lower temperatures facilitating the partitioning of ammonium nitrate into the particle phase. Generally, the MIR values are higher for episodes occurring in summer and early fall (EP1–EP4), whereas less difference in the PM mass is observed before and during episodes for the winter episodes (except for EP9), due to the higher background PM level. For the major PM constituents, MIR >1 was generally observed during all episodes, except during EP6N and EP10. Nitrate showed the largest MIR values, especially during EP1 and EP11N.

Increased formation of secondary inorganic aerosol during PM2.5 episodes has been extensively reported in the literature (e.g. Liu et al., 2020; Yun et al., 2018); thus, it will not be our focus. In the later sections, we will examine the chemical evolution of the precursor-specific SOA tracers during the episodes in detail.

Levoglucosan, originating from the pyrolysis of cellulose and hemicellulose, has been widely used as a BB POA tracer (Simoneit et al., 1999). In our dataset, a moderate correlation between 4-nitrocatechol and levoglucosan was observed (Rp of 0.43), which is in line with their common material origin from BB. The moderate correlation between the two species was not affected by meteorological factors (Fig. S14). It is also noted that 4-nitrocatechol was moderately correlated with benzene (Rp of 0.52) and toluene (Rp of 0.50), also implicating these anthropogenic VOCs as notable contributing precursors to 4-nitrocatechol. Furthermore, 4-nitrocatechol and NOx were moderately correlated (Rp of 0.48), suggesting the importance of NOx oxidation of the aromatic VOC precursors.

Seasonal variation in 4-nitrocatechol showed the highest concentration in winter (6.12±6.80 ng m-3), followed by fall (3.67±4.12 ng m-3) and summer (1.36±3.45 ng m-3). Levoglucosan showed comparable high concentrations in winter and fall (29.6±16.7 and 25.4±17.3 ng m-3 respectively), which were more than 2 times higher than that in summer (10.6±12.8 ng m-3). It is noted that the levoglucosan concentration in this study was lower than previous off-line filter-based measurements from fall to winter in 2010–2012 in HK (mean of 96.8 ng m-3), whereas the concentration of 4-nitrocatechol was comparable (mean of 3.42 ng m-3 in Chow et al., 2016). This likely reflects that 4-nitrocatechol has precursor sources other than BB (Lu et al., 2019; Yuan et al., 2021); thus, joint measurements of potential precursors (e.g. catechol, phenol, and benzene) in the future would help to discern the relative importance of precursors from BB vs. anthropogenic sources.

Figure 3a compares the average concentration of levoglucosan and 4-nitrocatechol for each pre-episode–episode time window pair. The variation in 4-nitrocatechol was generally in sync with the variation in NOx between the pre-episode and episode periods, signifying the importance of the NOx influence. Levoglucosan and 4-nitrocatechol jointly showed higher concentrations during most episodes, except for EP5 and EP6N, indicating BB as a frequent important contributor to episodic increases in PM2.5. Figure 3b shows the MIR distribution of levoglucosan and 4-nitrocatechol during each episode. MIR >1 was observed for both levoglucosan and 4-nitrocatechol during summer and early-fall episodes (EP1–EP4), with the latter showing noticeably larger MIR values (3–7). The observations suggested enhanced contributions from both primary BB emissions and BB-derived SOA, especially secondary formation during EP1–EP4. For EP5 and EP6N, the concentration of levoglucosan was higher in the pre-episode period compared with the episode hours. During winter episodes (EP7–EP11), MIR values were lower (<3), indicating lower differences between the pre-episode and episode hours. In the early-winter episodes (EP7–EP9), the MIR values of levoglucosan were higher than 4-nitrocatechol, suggesting a more important contribution from primary BB emissions during these episodes. In EP11N, a higher MIR value was observed for 4-nitrocatechol than for levoglucosan, and the increase in 4-nitrocatechol was accompanied by an increase in NOx, suggesting enhanced night-time secondary formation during this episode.

Two anthropogenic SOA tracers were measured in this study – phthalic acid and DHOPA. Phthalic acid is an SOA tracer from two-ringed naphthalene/methylnaphthalene (Kleindienst et al., 2012), whereas DHOPA is an SOA tracer from monoaromatics such as BTEX (benzene, toluene, ethylbenzene, and xylene) (Al-Naiema and Stone, 2017). Previous studies have shown that the main atmospheric loss of benzene and toluene is their photochemical reaction with OH radicals, with an atmospheric lifetime of 12.5 and 2.0 d respectively (Prinn et al., 1987). As there were few anthropogenic sources near the HKUST supersite, the measured BTEX is likely primarily due to transport from upwind mainland China, rather than local emissions, due to these species' long atmospheric lifetimes.

Phthalic acid showed a slightly higher concentration in fall and winter (16.8±12.5 and 15.3±11.0 ng m-3 respectively) and a lower concentration in summer (12.4±18.1 ng m-3). In comparison, DHOPA showed a more notable seasonal variation: the concentrations in fall and summer (1.17±1.09 and 1.01±1.50 ng m-3 respectively) were about 5 times the concentration in winter (0.26±0.22 ng m-3). VOC precursors of DHOPA (i.e. BTEX) showed the following pattern with respect to concentration: winter > fall > summer. Moreover, the seasonal average concentrations of these precursor species were 1.41±0.68, 0.97±0.58, and 0.68±0.86 ppb for the above-mentioned seasons respectively. Phthalic acid and DHOPA were positively correlated with O3 (Rp of 0.36 and 0.44), but DHOPA was not correlated with its aromatic VOC precursors (Rp of 0.03). The results suggest that the oxidant level is a significant factor promoting the formation of both phthalic acid and DHOPA, which is consistent with their secondary origin. Note that we use ozone as an indicator of the oxidant level in the ambient atmosphere, as no measurements of OH radical were available. The formation pathways for phthalic acid and DHOPA are mostly via OH radical oxidation, as reported in previous studies (He et al., 2018; Wang et al., 2007; Zhang et al., 2021).

Figure 4a shows the average concentration of phthalic acid and DHOPA quantified during the pre-episode and episode periods. Among the 11 episodes, higher concentration of the two SOA tracers were observed during summer and fall episodes (EP1–EP6), with episode-averaged values of 21.0–33.8 and 1.33–3.01 ng m-3 for phthalic acid and DHOPA respectively. During the winter episodes (EP7–EP11), the average concentrations of both SOA tracers were in a lower range, with DHOPA being significantly lower (0.10–0.36 ng m-3) and phthalic acid being slightly lower (8.59–20.4 ng m-3). As shown in Fig. 4b, BTEX had an enhanced presence during EP1–EP4, but no discernable elevation was detected during the remaining episodes in comparison with the pre-episode periods. DHOPA and phthalic acid also had higher MIR values during EP1–EP4 (MIR >2) than during the other episodes (MIR of 0.7–2), which is in line with the precursor–product relationship. The MIR values of DHOPA and phthalic acid exceeded unity during the remaining episodes, with EP7N being an exception for DHOPA and EP11N being an exception for phthalic acid. This seeming discrepancy in concentration variation between precursors and their product reflects that the key factors influencing formation of SOA tracers are not limited to their VOC precursors.

Three types of biogenic SOA tracers were quantified in this campaign, i.e. SOA tracers derived from isoprene, α-pinene, and β-caryophyllene. Previous studies have estimated that the respective atmospheric lifetimes of isoprene, α-pinene, and β-caryophyllene are 1.3 d, 4.6 h, and 2 min against ozone at a concentration of 7×1011 cm-3 (∼30 ppb) and 1.4 h, 2.6 h, and 42 min against OH at a concentration of 2×106 cm-3 (Atkinson and Arey, 2003). Thus, isoprene would preferentially react with OH, whereas β-caryophyllene would mainly react with O3. Compared with the precursors, the chemical lifetime of the SOA tracers is much longer, in the range of 2–10 d (Nozière et al., 2015). This implies that the SOA tracers observed at the site can be significantly contributed by regional/super-regional transport. The diurnal variation in the α-pinene and β-caryophyllene SOA tracers in this study showed clearly enhanced concentrations during the daytime from 10:00 to 16:00, whereas the isoprene SOA tracers did not show a discernable trend (Fig. S9).