As mentioned earlier, two emission-oriented hypotheses have been proposed as a possible cause of the O3 increases in 2015–2020. One is the changing emissions of the O3 precursors of NOx and VOCs (Li et al., 2022). The other is the reduced removal of HO2 radicals due to diminishing PM2.5 (K. Li et al., 2021; Shao et al., 2021). Li et al. (2022) showed that the NO titration effect was the cause of the linear trend in O3 in the PRD (0.5 ppb yr-1) during the relatively long period of 2006–2019. But for the period of 2015–2020, the NO titration effect could account for only about 10 % of the linear increase in O3 of the low-O3 stations in the PRD (5.0 ppb yr-1; green line; Fig. S3a).

The increase of 30 ppb in O3 at the low-O3 stations in BTH from 2015 to 2017 (green line; Figs. 4a and 8a) represents about 50 % increase in O3. The titration effect can account for only about 5 % (Fig. 8f). If this increase of 30 ppb in O3 were due to an enhancement in O3 precursors, then the enhancement would have to be substantially greater than 50 % because of the well-known less-than-linear relationship between the changes in O3 and its precursors; i.e., substantially higher percentage changes in precursors are needed for each percentage change in O3 (Dodge, 1977; Shafer and Seinfeld, 1985). Figure 8d and f show that CO (a proxy for VOCs) and NOx changed only by a few percent from 2015 to 2017, which is more than 1 order of magnitude less than the changes needed. Hence, it appears that changes in meteorological conditions conducive to O3 formation are more likely the major contributing factor to the 50 % increase in O3 at the low-O3 stations in BTH. A similar argument can be extended to the YRD and PRD (Figs. S1 and S3).

The theory of reduced removal of HO2 radicals by diminishing PM2.5 (25 %; green line in Fig. 8c) appeared to be qualitatively valid for the 50 % increase in O3 at the low-O3 stations in BTH from 2015 to 2017 (green line in Fig. 8a). But this theory was contradicted directly by the phenomenon at the high-O3 stations, where a 30 % reduction in PM2.5 (red line in Fig. 8c) corresponded to a decrease rather than an increase in O3 (red line in Fig. 8a).

Figure 9a shows the mean daily O3 concentrations of the first group with 4 or more consecutive O3-exceeding days (labeled O3 days ≥ 4) in 2015; Fig. 9b shows the mean daily O3 concentrations of the second group with fewer than 4 consecutive O3-exceeding days (labeled O3 days < 4); and Fig. 9c is the difference between the two groups (6.10 ppb; Table 2). Figure 9d–f are the same as Fig. 9a–c, respectively, but for 2017. The first group in 2017 had 28 d and mean O3 of 74.43 ppb inside the BTH box, while the second group had 34 d and 65.32 ppb (Table 2). One of the most remarkable differences between 2017 and 2015 in Fig. 9a–f was the large number of days with 4 or more consecutive O3-exceeding days (first group) in 2017 (28 d; Fig. 9d) over that of 2015 (7 d; Fig. 9a), which alone contributed to about 62 % of the difference in O3 between 2017 and 2015 as shown in Fig. 2a (red line). Approximately 30 % was contributed by the 10 d difference (2017 vs. 2015) in the number of days with fewer than four consecutive O3-exceeding days (second group). The contribution by the higher average concentration of MDA8 O3 of the first group in 2017 is only about 8 % (Table 2). These values of the contributions reconfirm what is shown in Fig. 3a; i.e., the greater frequency of episodes with 4 or more consecutive O3-exceeding days contributes the majority (62 %) to the higher O3 in BTH in 2017 vs. 2015, and the greater intensity/concentration of O3 during the episodes contributes only about 8 %, consistent with the expansion and saturation effect discussed earlier.

The phenomena illustrated in Fig. 9a–f also exist in the YRD and PRD, as well as in most other years. Figures equivalent to Fig. 9a–c for all years in the three city clusters are provided in the Supplement (Figs. S4–S6). Essential information derived from those figures is summarized in Tables 2–4.

In Fig. 10a and b, the values of SSR and T2m of the episodes with 4 or more consecutive O3-exceeding days are compared to those of O3 episodes with fewer than four consecutive O3-exceeding days and to those of clean days (non-O3-exceeding days). As expected, the O3 episodes with 4 or more consecutive O3-exceeding days consistently have the highest values of SSR and T2m, while the clean days have the lowest values. This is the case in nearly all years studied, as shown in the Supplement (Fig. S7), and is also generally true in the YRD and PRD regions (Figs. S8 and S9). Coupling the higher values of SSR and T2m in the O3 episodes with 4 or more consecutive O3-exceeding days depicted in Fig. 10 and a greater number of days in the O3 episodes with 4 or more consecutive O3-exceeding days shown in Fig. 3, we therefore propose a hypothesis as follows: the cause of worsening O3 increases in the BTH, YRD, and PRD regions from 2015 to 2020 could be attributed to enhanced photochemical O3 production due to increased occurrence of meteorological conditions of high solar radiation and positive temperature anomaly at the surface.

Quantitatively the coupling of Fig. 10 with Fig. 3 can be performed by multiplying the difference between the red (4 or more consecutive O3-exceeding days) and green (clean days) values of SSR or T2m in Fig. 10 with the frequency of occurrence (in percentage of total days) of O3 episodes with 4 or more consecutive O3-exceeding days from Fig. 3. The results are compared to the yearly O3-exceeding days in Fig. 11. The correlation between the yearly O3-exceeding days and weighed SSR is very good, with R values of 0.88 or greater in all three regions, lending strong support to our hypothesis. The correlation between the yearly O3-exceeding days and weighed T2m is highly correlated in BTH but not correlated in the YRD and PRD, which probably suggests that T2m is not as strongly coupled to O3 formation as SSR. Inclusion of O3 episodes with fewer than 4 consecutive O3-exceeding days in Fig. 11 did not change the correction coefficients significantly, supporting the robustness of results shown in Fig. 11.

Hu et al. (2023), in collaboration with this study, conducted a statistical analysis to assess the processes that contribute to high-O3 formation in the PRD when TCs were present in the northwestern Pacific. They investigated the impact of the distance between TCs in the northwestern Pacific and PRD on the O3 concentration in the PRD from 2006 to 2020. They found that the large numbers of consecutive O3-exceeding days in 2017 and 2019 relative to 2015 were primarily attributable to the greater occurrence of downdrafts and stable atmospheric conditions brought about by mid-distance-category TCs. This finding clearly establishes that changing frequency of mid-distance-category TCs (i.e., changing meteorological conditions) is the cause of the increases in the numbers of consecutive O3-exceeding days, as well as the higher-O3 concentrations in the PRD. An ongoing study by our research group further shows that the mid-distance-category TCs are predominately those TCs with tracks starting around the southern Philippines and ending near Korea and/or Japan. Since TC tracks in northwestern Pacific are strongly controlled by the WPSH, we conclude that both the Philippines-to-Korea or Japan track TCs and corresponding distribution and intensity of the WPSH contributed to the favorable meteorological conditions leading to higher consecutive O3-exceeding days in the PRD from 2015 to 2020.

Following the analysis by Hu et al. (2023), the mean vertical velocity at 850 hPa during all O3-exceeding days in the PRD in 2015 (Fig. 12a) is compared to that of episodes with 4 or more consecutive O3-exceeding days in 2017 (Fig. 12b). Major features in Fig. 12 compare very well with those of Fig. 7; e.g., the area with positive vertical velocity (downdrafts) in 2017 (red area in Fig. 12b), which was highly conducive to O3 formation, was by far more widespread and greater in value than that of 2015 (red area in Fig. 12a), agreeing better with the greater high-O3 area of Fig. 7b (2017) than that of Fig. 7a (2015). This agreement confirms that the increase in O3 in the PRD from 2015 to 2017 was caused by increased downdrafts and stable atmospheric conditions (meteorological conditions) brought about by TCs, as suggested by Hu et al. (2023). The same plots for BTH are shown in Fig. 13. The features of Fig. 13 are highly consistent with those of Fig. 5. The same plot for the YRD (Fig. S10) also showed more extensive and greater downdrafts in 2017 than 2015. However, the area of the positive vertical velocity in the YRD appeared to shift by about 500 km to the east compared to the area of high of O3 in Fig. 6b. Considering the uncertainty in evaluating the vertical velocity and that O3 formation is also dependent on parameters other than the vertical velocity, the discrepancy is acceptable.

To summarize this section, the increases in O3 in the three megacity clusters are critically dependent on the number of 4 or more consecutive O3-exceeding days. In addition, Hu et al. (2023) found that the changing frequency of mid-distance-category TCs (i.e., changing meteorological conditions) is the cause of the increases in the numbers of consecutive O3-exceeding days, as well as the O3 concentrations in the PRD. More importantly, our additional analyses of the mean vertical velocity at 850 hPa over the three megacity clusters (Figs. 12, 13, and S10) show that the increases in O3 in all three megacity clusters from 2015 to 2017 were caused by enhanced downdrafts and stable atmospheric conditions (meteorological conditions) which were highly conducive to O3 formation. The enhanced downdrafts and stable atmospheric conditions were brought about by TCs and the associated WPSH. Here we bring up the WPSH because it is well known that the tracks of TCs are influenced strongly by the WPSH and that WPSH strongly affects the regional atmospheric dynamics and therefore O3 formation (Chang et al., 2019; Mao et al., 2020; Ouyang et al., 2022; Zhao and Wang, 2017).

Mao et al. (2020) made a comprehensive study of an 11 d O3 episode in BTH in 2017 and found that it was dominated by the presence of the WPSH and mid–high-latitude wave activities. Depending on the position and intensity, the WPSH is well known to be a crucial factor affecting O3 concentrations in various parts of eastern China (Chang et al., 2019; Yin et al., 2019; Zhao and Wang, 2017). During this 11 d O3 episode, the ridge line of the WPSH maintained at approximately 22∘ N from 24 to 29 June 29, which, in combination with mid–high-latitude wave activities, induced meteorological conditions highly conducive to the O3 production in BTH and the northern YRD (Mao et al., 2020).

Following the analysis of Mao et al. (2020), the impact of the WPSH on O3 in BTH in April–September has been analyzed in Fig. 14, which depicts the composite 500 hPa geopotential height contours, humidity, and winds in BTH in April–September for O3-exceeding days in 2015 (Fig. 14a), clean days in 2015 (Fig. 14b), O3-exceeding days in 2017 (Fig. 14c), clean days in 2017 (Fig. 14d), O3-exceeding days in 2019 (Fig. 14e), and clean days in 2019 (Fig. 14f). The 3 years (2015, 2017, and 2019) are chosen because their differences in O3 contribute predominately to the overall O3 increases (Figs. 1–2). The importance of the WPSH is clearly visible in all Fig. 14a–f when the 5880 and 5900 gpm isolines (green lines) of O3-exceeding days are compared to those of clean days. In all 3 years, the WPSH of the former (O3-exceeding days) were significantly stronger than the latter (clean days), as evident by the strong anticyclonic winds and/or the larger areas inside the 5880 gpm isolines. Even in the case of 2017, when the area inside 5880 gpm isolines of the former looked to be similar to that of the latter, the appearance of the 5900 gpm line in the former indicated a stronger WPSH. The strong anticyclonic winds in the O3-exceeding days (Fig. 14a, c, and e) force the moist air of the South China Sea northward into southern China and contributed to extensive clouds and precipitation and thus low-O3 formation over southern China and the southern YRD. This difference in the O3 formation between BTH and southern China provides a good explanation for why the O3-exceeding days mostly occur in different time periods in the three megacity clusters, as discussed in Sect. 3.2. Furthermore, over the East China Sea, the prevailing westerlies were forced northward, slowed down, and led to meteorological conditions in BTH and the northern YRD characterized by a cloudless sky, a sinking motion, and high vertical stability in the lower troposphere, as well as high SSR and a positive T2m anomaly at the surface. These meteorological conditions were highly conducive to the formation and accumulation of O3. In contrast, the weaker WPSH of the clean days allowed relatively strong westerlies to prevail over BTH during clean days in the 3 years, which tended to disperse O3 (Fig. 14b, d, and f); e.g., the average wind speed over BTH was about 10 m s-1 in Fig. 14b, while it was only about 5 m s-1 in Fig. 12a. Quantitatively, Fig. 12c had 31 more O3-exceeding days (93 ppb) than Fig. 14a, but the 31 d came at the expense of clean days (52 ppb) (Fig. 14b and d). The contribution of these 31 d to the difference in MDA8 O3 between 2017 and 2015 (6.5 ppb; Fig. 1) can be calculated as follows: ((93×62)+(56×121))/(62+121)-((86×31)+(52×152))/(31+152)=10.8 ppb. This difference of 10.8 ppb in MDA8 O3 between 2017 and 2015 was for the period of April to September. It should be divided by 2 and then becomes 5.4 ppb for the yearly difference in MDA8 O3 between 2017 and 2015. This value of 5.4 ppb accounted for 83 % of the observed difference in MDA8 O3 between 2017 and 2015 (6.5 ppb; Fig. 1). A similar statement can be made for the difference in MDA8 O3 between 2019 and 2017 (Fig. 14e and c; Fig. 1).

We have made the same analysis for other years as well for the YRD and PRD. The results are mostly similar and thus presented in the Supplement (Figs. S11, S12, and S13). Figure 15a–d for the PRD in 2017 and 2019 is shown because there were interesting anticyclonic circulations over the PRD during O3-exceeding days in both years (Fig. 15a and c). The 2017 anticyclone was a direct product of the WPSH, as it resided within the western tip of the 5880 gpm isoline. The 2019 anticyclone was also likely associated with the WPSH, as the center of anticyclone resided just beneath the 5860 gpm isoline to the west of the PRD. The anticyclonic circulations were accompanied by stable downdrafts, low winds, and cloudless sky conditions (short arrows and blue shading in Fig. 15a and c), which were highly conducive to the O3 formation. Cloudless sky conditions also occurred in the YRD and BTH in Fig. 15a and c, but the high wind speed prevented the accumulation of O3. This difference in O3 accumulation between the PRD and other two regions provides another good explanation for why the O3-exceeding days mostly occur in different days in the three megacity clusters, as discussed in Sect. 3.2. Quantitatively, Fig. 15c had 27 more O3-exceeding days (90 ppb) than Fig. 13a, but the 27 d came at the expense of clean days (39 ppb) (Fig. 15b and d). The contribution of these 27 d to the difference in MDA8 O3 in the PRD between 2019 and 2017 (6.0 ppb; Fig. 1) can be calculated as follows: ((90×62)+(46×182))/(62+182)-((85×35)+(39×209))/(35+209)=11.6 ppb. This difference of 11.6 ppb in MDA8 O3 between 2019 and 2017 was for the period of April to November. It should be divided by 365/244 and then become 7.75 ppb for the yearly difference in MDA8 O3 between 2019 and 2017. This value of 7.75 ppb was 1.75 ppb more than the observed difference in MDA8 O3 between 2019 and 2017 (6.3 ppb; Fig. 1), suggesting a reduction of about 3.5 ppb in MDA8 O3 in the cold months of January–March and December between 2017 and 2019, which was approximately confirmed by the observed reduction of 3.88 ppb.

The presence of anticyclonic circulations over the PRD is in good agreement with the results of Ouyang et al. (2022) and Hu et al. (2023). The latter authors suggested that the anticyclonic circulations over the PRD were primarily caused by TCs in northwestern Pacific. Nevertheless, it is widely acknowledged that the tracks of TCs in the northwestern Pacific are influenced, at least to some extent, by the WPSH (Sun et al., 2015; Wang et al., 2017), making it difficult to separate the roles played by the TCs on the anticyclonic circulations and O3 formation from those of the WPSH. Clearly, further investigation is needed to fully understand the complex relationship among the WPSH, TCs, and O3. Based on these results, we hypothesize that the increased frequency of these meteorological conditions enabled by the changing intensity and position of the WPSH could contribute as a major cause of the O3 increases in the three megacity clusters in eastern China during 2015–2020.

Why do favorable meteorological conditions only cause an O3 increase at low-O3 stations but not at high-O3 stations? And why is the O3 saturation level at around 100 ppb, as shown in Fig. 4? These questions can be best addressed by examining Fig. 16, which depicts the time series of individual processes (where DDEP denotes dry deposition, CHEM the net photochemical production of O3, HTRA the horizontal transport, and VTRA the vertical transport) contributing to O3 budget in the PRD (averaged over 56 stations in the PRD) calculated by the WRF–CMAQ model for the O3 episode of 24 September–1 October 2019 (Ouyang et al., 2022). This episode was one of the most severe O3 episodes since the official O3 observation started in the PRD in 2006. MDA8 O3 exceeded the 75 ppb standard on all 8 d of the episode. Hourly O3 reached as high as 110 ppb, yet all MDA8 O3 stayed approximately between 75 and 100 ppb. This suggests a ceiling or saturation level of approximately 100 ppb for MDA8 O3, consistent with what was observed in Fig. 4 for the PRD, as well as BTH and the YRD. Since this episode was one of the most severe episodes, we can assume that the 100 ppb saturation level would also be applicable to other O3 episodes in Guangdong. More importantly, the saturation effect was also a common feature in the results of other three-dimensional models for other megacity clusters, in which MDA8 O3 usually saturated at 100 ppb, e.g., in the YRD (Li et al., 2012) and in Beijing (Zhang et al., 2023). This explains why the saturation O3 level is around 100 ppb, as shown in Fig. 4.

With regard to the first question above (why do the favorable meteorological conditions only cause O3 increase at low-O3 stations but not at high-O3 stations?), it can be understood as follows. At a low-O3 station of 65 ppb MDA8 O3 in the PRD in 2015 (Fig. 4c), Fig. 16 shows that MDA8 O3 can readily increase to 75–100 ppb in a few hours from an early morning low-ozone level of about 50 ppb under favorable meteorological conditions. However, at a high-O3 station of 100 ppb MDA8 O3 in 2015 (Fig. 4c) under the same favorable meteorological conditions, MDA8 O3 would also reach 75–100 ppb in a few hours from an early morning low-ozone level of about 50 ppb. (Note that we assume that all stations start the day with an early-morning low-ozone level of 50 ppb, consistent with the value in Fig. 16). In other words, the saturation levels at all stations are the same at 75–100 ppb, independent of the ozone concentration in 2015.

In terms of contributing processes, the saturation level of 75–100 ppb is controlled primarily by photochemical loss, dry deposition, and dispersion to the free troposphere. This can be clearly seen in Fig. 16 because on all 8 d in the mid-morning, when O3 is approaching toward its peak value, CHEM declines sharply due to photochemical loss and HTRA, VTRA, and DDEP all increase. Near noontime, O3 starts to drop sharply.