Correlation between air temperature and surface ozone in their extreme ranges in the greater Tokyo region

High-level ozone (O3) events observed around major urban regions in the middle latitudes are correlated with high temperatures (T-O3 correlation). Therefore, the effects of global warming on the future O3 levels are a matter of concern. The T-O3 correlation is caused by various physicochemical and meteorological processes, the importance of which can differ by region. This statistical analysis focused on the correlation in the extremely high ranges, because the lower ranges would only act as noise in elucidating the conditions at which high temperatures and high levels of O3 occur. This methodology was applied to the greater Tokyo region after 2001, where severe O3 events frequently occurred when the sea breeze system developed in summer. To select sample days for the analysis, this study set up twofold filtering: (1) a large threshold for midday sunshine duration and (2) a typical variation pattern to roughly judge sea breeze days, mostly essential weather pattern for high-level O3 events in the region. The most notable result was a decrease in O3 corresponding to the reduction in non-methane hydrocarbons (NMHC) from Period I (2001–2007) to Period III (2017–2019). As the NMHC rank reduced, the linear regression line for the T-O3 correlation shifted downward, but its slope (ppb/°C) remained around 10, except that temporary spikes in O3 levels and temperatures occurred at moderate NMHC levels. From an urban meteorological perspective, the wind speed in the mature stage of the sea breeze is the major factor behind the T-O3 correlation. • Daily maximum temperature-ozone correlation in a large urban region after 2001. • Temperature-O3 correlation focused on typical meteorological pattern in summer. • Reduction of O3 precursors not influences the slope of temperature-O3 correlation. • Sea breeze advance velocity is a major factor causing temperature-O3 correlation.


Introduction
Around the major urban regions located in the middle latitudes, high levels of ozone (O 3 ) frequently occur during the daytime in summer.Increases in ambient O 3 are caused by complex photochemical reactions of the precursors, mainly volatile organic compounds (VOCs) and nitrogen oxides (NO x ), emitted in the region.The reactions depend on solar radiation intensity.Simultaneously, the solar radiation affects various processes near the ground surface, particularly the increase in air temperature.Therefore, a positive correlation between air temperature and O 3 concentration (T-O 3 correlation) is generally observed.The T-O 3 correlation has led to concerns about the influence of higher temperatures on summer O 3 levels, which is associated with global warming.This problem was termed as the "O 3 climate penalty" and has been actively discussed (e.g., Kerr et al., 2019;Porter & Heald, 2019;Rasmussen et al., 2012).The main subject of these studies is the relative importance of various processes responsible for the T-O 3 correlation, because it does not represent a direct connection between them but is a synthetic result of many processes affected by solar radiation and temperature.For example, VOC emissions from many sources increase with temperature rise, which promotes O 3 production.Relative importance of the relevant processes can be different in every region, and no universal ranking is supposed to be significant.Therefore, area-specific research can provide valuable insights.
This study focused on the T-O 3 correlation observed in the greater Tokyo region and explored the changes in the correlation and their causes by using monitoring data after 2001.Dense and well-maintained monitoring data were obtained there.Notably, its T-O 3 correlation has not been investigated except a few brief reports (Yonemochi, 2019); thus, this study affords contribution to the literature.Before conducting the T-O 3 correlation analysis for a specific region, we fixed the methodology for the analysis.A discussion of the methodology is presented in Sect. 2. A brief introduction to tropospheric O 3 in the greater Tokyo region is as follows.
Tokyo and its downwind areas have long experienced severe air pollution.Since the 1970s, photochemical air pollution warnings have been issued most frequently in the country, and this region was the main target of countermeasures for reducing photochemical O 3 .Accordingly, high-level O 3 events in the region were studied repeatedly (e.g., Inoue et al., 2010;Kiriyama et al., 2015;Ueno & Tsunematsu, 2019;Wakamatsu et al., 1999).The basic structure and behavior of those events are localized within the Kanto Plain, although long-range transport from the Asian continent is occasionally observed in spring (Itahashi et al., 2009).The major sources of O 3 precursors are distributed in the southern part of the plain including the Tokyo metropolitan area.In contrast, the areas where O 3 reaches maxima extend from the source area to the northern part of the plain.This northward expansion of the pollution is almost solely brought by the southerly sea breeze prevailing during the daytime in the warm seasons (e.g., Kurita & Ueda, 1986;Wakamatsu et al., 1981).The sea breeze in this region is occasionally organized in a terrain-induced thermal low covering Central Japan (Kondo, 1990).Furthermore, light gradient winds from the Western Pacific high support the sea breeze to extend, although its effect is changeable.

Daily-base sampling
The selection of samples for analysis is a fundamental factor in this methodology.Daily sets of reference values for various properties are mostly basic samples.Long-term data are often organized as monthly or annual averages.However, averaged datasets do not necessarily provide suitable features.For example, Ooka et al. (2011) demonstrated a strong correlation between the average daily maximum temperature (T max ) and maximum O 3 concentration for the summer months of every year between 1985 and 2005.However, these averaged values can be mutually correlated through their dependences on the number of clear-sky days (cloud cover less than a fixed value), even if high temperatures and O 3 levels do not occur.Specifically, the correlation obtained from yearly averaged data does not purely reflect physicochemical processes (Yoshikado, 2021).Changes in future O 3 level that can occur with warming should be examined as a result of physicochemical processes on a daily basis rather than long-term average properties.Therefore, the analysis used samples from a daily dataset.

Selecting clear-sky days
The most important factor of temperature and O 3 production is solar energy.It can be measured by sunshine duration (SS) or, more precisely, solar radiation intensity.The T-O 3 correlation is a composite of T-and O 3 -dependences on SS.Accordingly, any variation in T and O 3 , irrespective of the abundance of SS, can disturb the T-O 3 correlation.In addition to meteorological conditions, seasonal transitions such as changes from early summer to mid-summer, spatiotemporal variations in precursors, and other factors can cause such disturbances.In other words, even during hot seasons when severe O 3 events frequently occur, the relationship between temperature and O 3 level under cloudy-sky conditions should be considered to have a different system from that on clear-sky days.Because the response of the O 3 level to climatic warming is of interest, focusing on the T-O 3 correlation in the high-temperature range is a reasonable choice.Restriction of statistics to "clear-sky days" is also a suitable choice.Consequently, most of our analyses applied a threshold to the SS of the sample days.Detailed data information is given in Sect.3.

Selecting sea breeze days
Diversity of wind systems prevalent in the region can be a major factor disturbing the T-O 3 correlation.In other words, different wind systems contribute to each specific T-O 3 correlation.In coastal source areas, onshore gradient winds generally tend to lower the temperature and raise the inland-side O 3 concentrations, whereas offshore gradient winds raise temperature and lower O 3 .Specifically, in the greater Tokyo region, onshore sea breezes prevail on summer days (e.g., Kurita & Ueda, 1986;Yoshikado, 2013), and high-level O 3 events typically occur under those conditions (Khiem et al., 2010).Therefore, as a final step, days on which the sea breeze system satisfied a fixed definition, and the lowest limit of SS was exceeded as mentioned in the foregoing subsection, were selected as samples for our analysis.Details of the applied conditions are described in Sect.3.3.

Region and monitoring sites
The metropolitan area of Tokyo is located along the northwest corner of Tokyo Bay, and greater Tokyo extends northwest and along the bay shore (Fig. 1).Photochemical O 3 is generally produced in the sea breezes developing from Tokyo and Sagami bays.These sea breezes vary in intensity and extent under the influence of a synoptic-scale pressure gradient that decreases northward and the heat-island effect of the metropolitan area (Yoshikado, 1992).Thus, O 3 maxima are observed mostly within the urban areas or northern suburbs.The eastern parts of the Tokyo and Saitama prefectures (cf. Figure 1) correspond to the highest O 3 zone, which can be arbitrarily divided into S, M, and N sub-areas.For determining the maximum level in the region, 18 representative O 3 monitoring sites were selected.The maximum "level" was obtained as the average of the first and the second largest (hereinafter referred to as Top 2) daily maximum 1-h O 3 records from the 18 sites because a peak value from a single monitoring site might simply represent localized irregularities.
The main sources of O 3 precursors are distributed in and around the metropolitan area (sub-area S), where 13 monitoring sites measuring hourly non-methane hydrocarbons (NMHC, used as a substitute for VOCs) and NO x are in operation.The average NMHC and NO x of 13 sites were used in this study.
The daily maximum hourly record of T from the Saitama-Urawa automated station (AMeDAS) was employed as a representative.Hourly SS and wind (speed and direction) data were sourced from the Saitama-Urawa and Tokyo AMeDAS.SS was used to select clearsky days and in the process selecting sea breeze days.Detailed judgment flows are presented in Sect.3.3.

Data periods analyzed
Since our prime concern was the increase in O 3 on hot summer days, all 62 days in July and August of every summer were selected as the first stage.June was excluded Fig. 1 Monitoring sites used in the study as representatives of O 3 in South (S, eastern Tokyo), Middle (M, south-eastern Saitama), and North (N, North-eastern Saitama) sub-areas, and the automated meteorological sites (AMeDAS) in Tokyo (T) and Saitama-Urawa (U).Thirteen monitoring sites, from which average levels of NMHC and NO x were obtained, were all in the S sub-area because it is a rainy month in Central Japan.April and May were also used for reference because the weather in the region is generally fair and high-level O 3 events occasionally occur even without very high temperatures in these spring months.March has a considerably different climate, and hence, it was excluded here.Relevant data from 2001 to 2019 were available to examine recent trends.
The 3-h averages of the O 3 precursors during 0600-0900 JST (hereafter referred to as 06-09 h) were used as reference values to measure the effect on O 3 production until midafternoon.The recent trends are shown in Fig. 2. Both the NMHC and NO x decreased overall, specifically, in a stepwise manner, with relatively flat curves during Periods I and II.
Variations in precursors influence the relationship between T and O 3 .The ratio NO x /NMHC (Fig. 3) increased between Periods I and II as a result of the effective reduction in NMHC compared to that in NO x .According to Inoue et al. (2019), the O 3 formation regime in the Tokyo metropolitan area was VOC sensitive during the early stages of this century.Therefore, the conspicuous reduction in NMHC (in place of VOC in this study) might have been a significant countermeasure to high levels of O 3 , as confirmed in Sect.4.3.Wakamatsu et al. (2013) provided a historical review of these measures.
However, based on the regression lines in Fig. 3, the ratio NOx/NMHC changed only within a range of 0.13 to 0.17, which can be said no substantial variation.For reference, Sakamoto et al. (2019) reported the ratio at which the O 3 production regime switches to be approximately 0.067 (the inverse ratio 15).Therefore, the ratio range in this study was sufficiently far from the regime transitional zone, and only the NMHC data are referred to in the following section of the analysis.Periods I and II were selected for the correlation analysis.Additionally, Period III was examined to determine the most recent situation, although this period was shorter than the others.

Clear-sky days and sea breeze days
In accordance with the methodology stated in Sect.2, the "clear-sky days" were selected from the whole summer days (partially with spring days) in the three periods for further analysis.A "clear sky" was defined as follows: SS (expressed by the rate of 0-10 for every hour) reached or All days in the periods were classified into two groups: sea breeze (SB) days and non-SB days, to characterize atmospheric transport in the region.Only the SB days with SS6h ≥ 45 were considered for the T-O 3 correlation analysis.Here, a composite of several conditions was applied to identify SB days.It was originally established by Yoshikado (2013) and is described as follows: i) SS6h ≥ 30, not as large as the threshold for clear-sky days ii) The wind at the Tokyo site was northerly (WSW-E clockwise), or less than 4.5 ms −1 during 04-06 h, and item (iii) did not start before 08 h.iii) The wind changed to southerly (ESE-SW clockwise) between 08 and 14 h and persisted for two consecutive hours or longer at a speed faster than 2 ms −1 .iv) At the Urawa site, conditions similar to (i)-(iii) for the Tokyo site were satisfied, but the onset of the southerly wind in (iii) was not earlier than that at the Tokyo site.
Conditions (i)-(iv) reflect the empirically established fact that the sea breeze system tends to develop under clear skies with light gradient wind, but not with predominant gradient winds, even under clear skies.

Reference values characterizing sample days
There could have been arbitrariness in finding reference values that represented each sample day in the analysis.Meteorological factors T and SS were generally slightly localized in representing daily conditions in the region, and two AMeDAS stations were selected as indicated above.The daily maximum T (T max , precisely, largest hourly T) near the regional center (U in Fig. 1) was employed for the T-O 3 correlation analysis.Although an average T during the daytime might be more appropriate to connect with daily O 3 production, T max and average T were correlated with a correlation coefficient R > 0.95.Data representing the air pollutants are presented in Sect.3.1.Additional comments regarding the maximum level of O 3 are provided below.
The area where high-level O 3 events occur is generally changeable, even within a day and from day to day, in the region shown in Fig. 1.As a reference value of O 3 , "Top 2" from the whole region was employed (Sect.3.1).A preliminary examination compared the Top 2 and the maximum daily 8-h average O 3 (MDA8), a frequently used reference, resulting in a high correlation (R larger than 0.95) for the entire region.In the comparison at each station, R tended to be smaller.

Sample groups during three periods
In the first stage, four sample sets consisting of clearsky SB days were obtained from Periods I (2001-2007) and II (2010)(2011)(2012)(2013)(2014)(2015)(2016), as shown in Table 1.Although SB days occurred roughly every 3 days in both spring and summer, clear-sky condition (SS6h ≥ 45/60) reduced the final number of samples, especially in summer.Specifically, even in SB cases, cloudy skies partially extended more frequently in summer.The sample size was not sufficiently large for statistical analysis.To supplement the summer samples, especially considering the more recent circumstances in the study, the fifth sample set from Period III (2017-2019) was added later, although this period was shorter than that of the others.

Overview of Periods I and II
The T-O 3 correlations for Periods I and II are shown in Fig. 4. Overall, the daily maximum levels in O 3 (Top 2) increased with the daily maxima in T and were lower in Period II than in Period I. Dispersions were fairly small as a result of twofold screening by SS and the wind system (SB).The spring and summer samples lay together in series, although their T ranges were mostly different.
While the O 3 plots lay smoothly along the linear regression lines in spring (slope 4.53 and 3.31 ppb/°C), those in summer had steeper lines.The variances in summer appeared to be very large owing to their narrow T ranges.
When the summer plots in Fig. 4 were observed on a wider horizontal axis, there was a phase shift at approximately 33 °C; the samples exceeding 33 °C had much steeper regression lines than those in spring, whereas those below 33 °C acted similarly to the spring samples.
Therefore, the high T range (≥ 33 °C) should be further examined.

Response to the reduction in O 3 precursors
The large variances in the T-O 3 correlations for the summer cases resulted from the narrow T range between 33 and 38 °C, and wide variation in O 3 , even within the set of clear-sky SB days.The primary factor dominating O 3 production was the precursor level at the initial stage of O 3 production.This is largely influenced by local meteorology (Yoshikado, 2015).
The T-O 3 correlation of the part T ≥ 33 °C is shown again in Fig. 5 after regrouping by four ranks (cf.Table 2) of the NMHC reference value (Sect.3.2).All sample days of rank 4 occurred in Period I, and most of the rank 2 samples occurred in Period II.Rank 3 occurred during Periods I and II, representing a transition between these two periods.As indicated by the relative positions of the regression lines, the O 3 level decreased with an increasing NMHC rank.However, the slope for rank 2 (Fig. 5b), 17.5 ppb/°C, was abnormally larger than the others.This change can be traced to several events of high O 3 exceeding 150 ppb in the highest T range that occurred in Period II.Specifically, the O 3 level in this T range did not decrease conspicuously along with the decrease in the NMHC.This may have been associated with climatic fluctuations over a short period.Further discussion is provided later in connection with Fig. 6a.
In Period II, a few sample days fell into rank 1.Therefore, those in Period III were supplemented with the most recent summers, as shown in Fig. 5c.The regression line for these data tended to be parallel (slope approximately 10 ppb/°C) to those of ranks 3 and 4 but not to that of rank 2, although only few sample days distributed in the highest T range.
Even after sub-grouping the samples by NMHC rank, the variances in O 3 were somewhat large, and the correlation coefficients (R) were not sufficiently large: from ranks 4 to 1, R = 0.51, 0.37, 0.56, and 0.53.According to a detailed analysis of sample days that largely deviated from the regression line (Yoshikado, 2021), their wind patterns were quite irregular, even though they satisfied the sea breeze criteria described in Sect.3.3.The main point of irregularity was the large delay of sea breezes in reaching the inland site of Urawa.
The O 3 values at 35 °C, the approximate middle point on the regression lines in Fig. 5, were, from the highest NMHC rank, 167, 142, 125, and 104 ppb.Specifically, reductions in the NMHC rank of every 0.1 ppmC interval with roughly parallel reductions in NO x caused a reduction in O 3 of approximately 20 ppb.

One major factor responsible for the T-O 3 correlation
As mentioned above, the variance in O 3 in Fig. 5 appears to be magnified by cases of late inland penetration of sea breezes.Considering that slow penetration mostly corresponds to a small SB velocity in the coastal area, the wind speed at the Tokyo site averaged for four hourly records of 13-16 h (W ave 13-16) was selected as representative.
Figure 6 shows a re-examination of Fig. 5 for NMHC ranks 1 and 2 separated by the SB velocity at 4.5 ms −1 .At lower wind speeds (Fig. 6a), the regression line was lifted by very high levels of O 3 , although the variance was large.When Fig. 5b is compared with Figs.6a and b, most sample days above 160 ppb of O 3 are found in the lower velocity case (Fig. 6a).Furthermore, deeper inspections revealed that these sample days tended to occur in succession within several days as well as during specific summers.This suggests that the large slope for NMHC rank 2 (Fig. 5b) could be associated with climatic fluctuations characterized by low SB velocities.Further analysis is required to determine the effects of short-period climate (typically associated with ENSO) on O 3 trends.
Figure 7 shows how sea breeze intensity affects the extreme increases in T and O 3 .The symbols indicate sub-areas S, M, and N (cf. Figure 1) where O 3 exceeded 120 ppb for two successive hours at two or more of the six sites.The general tendencies are as follows.
i) The occurrence rate of high-level O3 events above 120 ppb (colored symbols) is high at lower wind velocities; in contrast, that of low-level O3 (symbol x) is high at higher wind velocities.ii) High temperatures in the extreme range occur at low wind velocities and are accompanied by high-level O 3 events in most cases.iii) High-level O 3 events at low wind velocities mostly occur in the S sub-area where the urban center is located (in addition to S, symbols SM and SMN indicate its occurrence in the S sub-area).Even under intense sea breezes, O 3 occasionally reached high levels in the N sub-area.
The factors enhancing summer heat around a large urban region presumably aid in understanding the T-O 3 correlation in the highest temperature range.Yoshikado (1992) provided a perspective from urban meteorology regarding the behavior of the sea breeze in the greater Tokyo region.Generally, the southerly gradient wind cooperates with the sea breeze to pass through the urban center by noon or early afternoon.However, under a light gradient condition, the urban heat island along the coast prevents the sea breeze from passing through, during which a stagnant area forms over the northern suburbs.This stagnant condition intensifies warming and simultaneously causes the accumulation of air pollutants in the urban and northern suburban regions.Findings (i)-(iii)  Although the contribution of stagnation to the T-O 3 correlation has been discussed as an important factor (e.g., Coates et al., 2016;Garrido-Perez et al., 2019;Pusede et al., 2015), no specific meteorological structure has been described.The present case establishes the behavior of the sea breeze as a major factor responsible for the T-O 3 correlation over a single heavily urbanized region.

Discussion and conclusion
The T-O 3 correlation has been broadly recognized.Therefore, the effects of global warming on future O 3 levels are concerning.In further research, investigations of the mechanisms responsible for intense heat in selected regions, and temperature-dependent factors raising O 3 levels, are required.As a distinctive methodology, the present study focused on the T-O 3 correlation for summer days within specific ranges of T max and SS and furthermore with specific wind patterns.
In the greater Tokyo region, the daytime peak of O 3 decreased as the emissions of the O 3 precursors decreased, except for temporary persistence recorded under very hot conditions in Period II (2010)(2011)(2012)(2013)(2014)(2015)(2016).A tentative interpretation of this exception was given in Sect.4.4.In the most recent period, Period III (2017-2019), the precursors were further reduced to the NMHC rank 0.05-0.15ppmC in most cases, and no irregularly high O 3 was found in the very high-temperature range.Therefore, the T-O 3 correlation can be summarized generally as the slope of the regression line of approximately 10 ppb/°C over a considerably wide range of variations in the precursors since 2001.These large slope values resulted from the daily maximum in O 3 and the daily maximum temperature on clear-sky days, ranging within 33-38 °C, and were not comparable with those based on the maximum daily 8-h average O 3 (MDA8) and nonspecific weather conditions (e.g., Bloomer et al., 2009;Ninneman & Jaffe, 2021).Because the ratio of MDA8 to daily maximum was approximately 0.7 from the preliminary analysis, the slope on the MDA8 base could be approximately 7 ppb/°C, still larger than those in literature, such as 0-6 (Kerr et al., 2019).If T max on clear-sky summer days rises owing to global warming, the large slope of the regression of the T-O 3 correlation in the high-temperature range examined in this study will be of significance.However, Otero et al. (2021) reported that the slope decreases with the reduction in O 3 precursors.Furthermore, the inverse view that the O 3 level does not rise much more in a higher temperature range (e.g., Steiner et al., 2010) needs examination.The realities of these findings depend on future data accumulated in various regions.
In addition to the precursors level dominating O 3 production, this study focused on the contribution of local meteorology to O 3 increase, major factor of which was the sea breeze intensity.The southerly gradient wind superposed on the sea breeze; therefore, the behaviors of the Western Pacific high as its source, are dominant factors.Fine weather in summer (typically, SS6h ≥ 45) is also accompanied by this high-pressure system.Therefore, if the summertime predominant pressure distribution changes, its prediction will be a key point when considering its dominant influence on the sea breeze behaviors.
The present data suggest that the frequency of high levels of O 3 will be low as long as the decreasing trend in the precursors is maintained.However, we cannot predict about the situation where "very hot days" frequently occur due to global warming because such sample days hotter than 38 °C are still rare.Another limitation of the present analysis is that the precursors were represented by NMHC data.Ueno and Tsunematsu (2019) reported that the total NMHC reactivity in O 3 production was unchanged in the Tokyo metropolitan area between 1990 and 2010 despite its changeable composition.For estimating future O 3 levels, changes in the composition of VOCs in the ambient air and their reactivity should be considered.In addition, the sensitivity of O 3 production to precursors must be considered in future reduction measures.It was found to be VOC-sensitive in the target region, and the reduction in NMHC successfully proceeded with NO x reduction, with a similar trend.However, their future changing rates might shift to NO x sensitive or mixed sensitive and further influence the T-O 3 correlation pattern.

Fig. 2
Fig. 2 Trends in the reference values of the O 3 precursors in summer (averages of all the available 13 monitoring sites for NMHC, where NO x is also measured).Cases denoted by "SB" indicate averages for exclusive sea breeze days and the others for the whole 62 days

Fig. 4 a
Fig. 4 a T max vs. daily maximum O 3 (Top 2) for Period I.The clear-sky SB days in spring and summer are plotted with different colors.Linear regression lines for spring, summer, and cases T max ≥ 33 °C in summer are shown along with the respective statistics.b Same as in (a) except for Period II

Fig. 5 T
Fig. 5 T max vs. daily maximum O 3 (Top 2) for the section T max ≥ 33 °C after grouping by four ranks of the NMHC reference value.a NMHC rank 4 (0.35-0.65 ppmC), occurred only in Period I, and rank 3 (0.25-0.35 ppmC) in Periods I and II.Regression lines correspond to the rank of the same color symbol.b NMHC rank 2 (0.15-0.25 ppmC).c NMHC rank 1 (0.05-0.15 ppmC)

Fig. 6
Fig. 6 Rearrangement of Fig. 5b and c with respect to the SB velocity.a W ave 13-16 ranged within 3-4.5 ms −1 and b ranged within 4.5-6 ms −1 .Here, the regression analysis data are from the whole T max range

Fig. 7 T
Fig. 7 T max vs. W ave 13-16 on the clear-sky SB days (NMHC ranks 1 and 2).Symbols indicate the sub-area(s) where high-level O 3 events occurred

Table 1
Number of days included in each category

Table 2
Number of sample days *1 (SS6h ≥ 45, SB days) classified to four ranks of the NMHC reference value *2 with the correlation coefficient (R) and the slope of regression (S) 3 ) The upper limits are not included (i.e., less than) *