Observed trends in ground-level O3 in Monterrey, Mexico, during 1993–2014: comparison with Mexico City and Guadalajara

Here, we present an assessment of long-term trends in O3 and odd oxygen (O3+NO2) at the industrial Monterrey metropolitan area (MMA) in NE Mexico. Diurnal amplitudes in Ox (AVd) are used as a proxy for net O3 production, which is influenced by the NO2 photolysis rate. No significant differences in the AVd are observed between weekends and weekdays, although the largest AVd values are observed at sites downwind of industrial areas. The highest O3 mixing ratios are observed in spring, with minimum values in winter. The largest annual variations in O3 are typically observed downwind of the MMA, with the lowest variations generally recorded in highly populated areas and close to industrial areas. A wind sector analysis of mixing ratios of O3 precursors revealed that the dominant sources of emissions are located in the industrial regions within the MMA and surrounding area. Significant increasing trends in O3 in spring, summer, and autumn are observed depending on site location, with trends in annual averages ranging between 0.19 and 0.33 ppbyr−1. Overall, from 1993 to 2014, within the MMA, O3 has increased at an average rate of 0.22 ppbyr−1 (p < 0.01), which is in marked contrast with the decline of 1.15 ppbyr−1 (p < 0.001) observed in the Mexico City metropolitan area (MCMA) for the same period. No clear trend is observed from 1996 to 2014 within the Guadalajara metropolitan area (GMA).


Introduction
O 3 is a secondary air pollutant formed in the troposphere via the photo-oxidation of CO, methane (CH 4 ), and volatile organic compounds (VOCs) in the presence of NO and NO 2 (NO+NO 2 = NO x ) (Jenkin and Clemitshaw, 2000).The system of O 3 production is not linear and is termed NO x -limited, when O 3 production increases in response to increasing NO x emissions and termed VOC-limited when it responds positively to emissions of VOCs (Monks et al., 2015;Pusede et al., 2015).Tropospheric O 3 is of concern to policy makers due to its adverse impacts on human health, agricultural crops, and vegetation and also due to its role as a greenhouse gas despite its relatively short lifetime of around 22.3 ± 3.0 days (Stevenson et al., 2006;IPCC, 2013;WHO, 2014;Lelieveld et al., 2015).As the predominant source of OH, tropospheric O 3 controls the lifetime of CH 4 , CO, and VOCs, among many other air pollutants (Revell et al., 2015).In polluted regions, increased levels of O 3 are prevalent during seasons with stable high-pressure systems and intense photochemical processing of NO x and VOCs (Dentener et al., 2005;Xu et al., 2008) with downward transport from the stratosphere of lesser importance (Wang et al., 2012).By contrast, the main removal processes for tropospheric O 3 are chemical loss and dry deposition (Atkinson, 2000;Jenkin and Clemitshaw, 2000).
Tropospheric O 3 increased in the Northern Hemisphere (NH) during 1950-1980 due to rapid increases in precursor emissions during the industrialisation and economic growth of Europe and North America (NA) (Staehelin and Schmid, Published by Copernicus Publications on behalf of the European Geosciences Union. 1991; Guicherit and Roemer, 2000).Since the 1990s, reductions in O 3 precursor emissions in economically developed countries have resulted in decreases in tropospheric O 3 levels (Schultz and Rast, 2007;Butler et al., 2012;Pusede et al., 2012); however, in some regions, increases in O 3 have also been reported.For instance, from an analysis of O 3 data from 179 urban sites over France during 1999-2012, Sicard et al. (2016) reported an increasing trend in the annual averages of 0.14 ± 0.19 ppb yr −1 , and in the medians of 0.13 ± 0.22 ppb yr −1 , attributed to long-range transport and reduced O 3 titration by NO due to reductions in local NO x emissions.However, Sicard et al. (2016) also reported during the same period that at 61 rural sites, O 3 decreased in the annual averages by 0.12 ± 0.21 ppb yr −1 and in the medians by 0.09 ± 0.22 ppb yr −1 .
In the US and Canada, O 3 levels have decreased substantially at different metrics during the last 2 decades in response to more stringent emissions controls focused on onroad and industrial sources.In the greater area of Toronto from 2000 to 2012, O 3 levels decreased at urban sites by approximately 0.4 % yr −1 , and at suburban sites by approximately 1.1 % yr −1 , as a consequence of a reduction in the midday averages of NO 2 of 5.8-6.4 % yr −1 and in the VOC reactivity of 9.3 % yr −1 (Pugliese et al., 2014).Emissions estimates suggest an overall national scale decrease during 1980-2008 in US NO x and VOCs emissions of 40 and 47 %, respectively, with city-to-city variability (EPA, 2009;Xing et al., 2013).Lefohn et al. (2010) reported that for 12 US major metropolitan areas, the O 3 US EPA exposure metrics of the annual second highest 1 h average and the annual fourth highest daily maximum 8 h average decreased during 1980-2008 at 87 and 71 % of the monitoring sites evaluated, respectively.However, Lefohn et al. (2010) observed an increase in the lower-and mid-O 3 mixing ratios in response to decreased titration by NO.More recently, Simon et al. (2015) assessed changes in the 1 h average O 3 mixing ratios at around 1400 sites across the US between 1998 and 2013, using the 5th, 25th, 50th, 75th, and 95th percentiles and the maximum daily 8 h average.Overall, Simon et al. (2015) observed increases at the lower end of the O 3 data distribution of 0.1-1 ppb yr −1 , mostly in urban and suburban areas, whereas O 3 decreased at the upper end of the data distribution between 1 and 2 ppb yr −1 in less urbanised areas.Such changes were associated with the implementation of control strategies within the US, such as the NO x SIP call, and tighter point and vehicle emissions standards to abate peak O 3 mixing ratios.
In Mexico, studies of long-term trends in O 3 have focused on the Mexico City metropolitan area (MCMA) (Molina and Molina, 2004;Jaimes et al., 2012;Rodríguez et al., 2016), with reports of a decrease in O 3 annual averages of ca.33 % during the last 2 decades (Parrish et al., 2011;SEDEMA, 2016a).O 3 has received less consideration in other large metropolitan areas, where Mexican air quality standards are frequently exceeded (Table 1).Indeed, since 2000, recorded O 3 mixing ratios have exceeded Mexican official standards for 1 h average (110 ppb) and 8 h running average (80 ppb) O 3 by more than 50 % in the Guadalajara metropolitan area (GMA, the second most populated city) and in the Monterrey metropolitan area (MMA, the third most populated city) (INE, 2011;SEMARNAT, 2015).To date, only Benítez-García et al. (2014) have addressed changes in ambient O 3 in the GMA andMMA during 2000-2011, reporting an increase in O 3 annual averages of around 47 and 42 %, respectively.However, it should be noted that the ordinary linear regression analysis used by Benítez-García et al. (2014) may be biased by extreme values and is therefore not suitable to determine long-term O 3 trends with significant confidence.
To improve air quality, the Mexican government has introduced several initiatives to reduce primary pollutant emissions, with emissions estimates reported in the Mexican National Emissions Inventory (NEI).The NEI suggests that from 1999 to 2008, anthropogenic NO x emissions decreased in the MCMA by 3.8 % yr −1 , but increased in the GMA and the MMA by 1.9 and 4.0 % yr −1 , respectively (Fig. S1 in the Supplement) (SEMARNAT, 2006(SEMARNAT, , 2011(SEMARNAT, , 2014)).These NEI NO x emissions estimates agree with the decrease for the MCMA of 1.7 % yr −1 in the NO 2 vertical column density during 2005-2014 reported by Duncan et al. (2016), but disagree for the GMA and the MMA where decreases of 2.7 % yr −1 and of 0.3 % yr −1 , respectively, are reported.Similarly, Boersma et al. (2008) observed that NO x emissions over Mexico derived from NO 2 satellite observations were higher by a factor of 1.5-2.5 times than bottom-up emissions estimates, which were lower by a factor of 1.6-1.8than data reported in the NEI 1999 base year.The NEI anthropogenic VOC emissions estimates suggest a decrease in the MMA by 0.2 % yr −1 but increases in the MCMA and in the GMA by 2.7 % yr −1 and by 3.2 % yr −1 , respectively (Fig. S1) (SEMARNAT, 2006(SEMARNAT, , 2011(SEMARNAT, , 2014)).However, as for NO x , NEI trends in VOCs disagree with existing reports for average VOC decreases within the MCMA (Arriaga-colina et al., 2004;Garzón et al., 2015).
Local authorities have developed local emissions inventories for the MCMA and the MMA, although only for the MCMA have the inventories been compiled with a frequency of 2 years since 1996 (SEDEMA, 1999(SEDEMA, , 2001(SEDEMA, , 2003(SEDEMA, , 2004(SEDEMA, , 2006(SEDEMA, , 2008(SEDEMA, , 2010(SEDEMA, , 2012(SEDEMA, , 2014(SEDEMA, , 2016b;;SDS, 2015).The accuracy of the MCMA emissions inventories has also been assessed during several field campaigns.For instance, during the MCMA 2002-2003campaign, Velasco et al. (2007) observed an overestimation in the 1998 inventory for VOC emissions of alkenes and aromatics but an underestimation in the contribution of some alkanes.By contrast, for the 2002 MCMA inventory, Lei et al. (2007) reported an underestimation in the VOC total emissions of around 65 %, based on a simulation of an O 3 episode that occurred in 2003 within the MCMA.Therefore, since these emissions estimates are used to predict future air quality and to design clean air poli-  cies, it is imperative to examine the results of the policies implemented to control emissions of O 3 precursors.
To our knowledge, no previous study has addressed trends in O 3 and odd oxygen in urban areas of Mexico.In this study, we describe trends in ground-level O 3 within the MMA and its response to changes in precursor emissions during 1993-2014.Long-term and high-frequency measurements of O 3 were recorded at five air quality monitoring stations evenly distributed within the MMA.In order to better assess photochemical production of O 3 , odd oxygen defined as ) was also considered, as O 3 and NO 2 are rapidly interconverted.Diurnal and annual cycles of O 3 and O x are used to interpret net O 3 production within the MMA.We show that air mass origin strongly influences the O 3 annual increases.The trends in O 3 , O x , and precursor emissions are compared with those observed within the MCMA and GMA.Finally, we describe how NEI emissions estimates for NO x and VOCs disagree in the trend magnitudes with ground-based NO x and VOCs measurements made in the urban areas studied here.
This paper is organised as follows: Sect. 2 presents the data quality and methodology used to derive the different trends presented.Section 3 describes in detail the O 3 and O x diurnal and annual cycles and annually and seasonally averaged trends.Section 4 discusses the origin of the O 3 and O x diurnal variations and trends in the light of changes in precursor emissions.Finally, Sect. 5 provides some conclusions regarding the trends observed in the urban areas studied.

Monitoring of O 3 in the Monterrey metropolitan area (MMA)
The MMA (25 • 40 N, 100 • 20 W) is located around 720 km N of Mexico City, some 230 km S of the US border in the state of Nuevo León (Fig. 1a).It lies at an average altitude of 500 m a.s.l. and is surrounded by mountains to the S and W, with flat terrain to the NE (Fig. 1b).The MMA is the largest urban area in northern Mexico at around 4030 km 2 and is the third most populous area in the country with 4.16 million inhabitants, which in 2010, comprised 88 % of the population of Nuevo León State (INEGI, 2010).It is the second most important industrial area in Mexico and has the highest gross domestic product per capita (Fig. 1c).Although the weather changes rapidly on a daily timescale, the climate is semi-arid with an annual average rainfall of 590 mm and an annual average temperature of 25.0 • C with hot summers and mild winters (ProAire-AMM, 2008;SMN, 2016).Within the MMA, tropospheric O 3 , six additional air pollutants (CO, NO, NO 2 , SO 2 , PM 10 , and PM 2.5 ), and seven meteorological parameters (wind speed (WS), wind direction (WD), temperature (Temp), rainfall, solar radiation (SR), relative humidity (RH), and pressure) have been monitored continuously, with data summarised as hourly averages, since November 1992 at five stations that form part of the Integral Environmental Monitoring System (SIMA) of the Nuevo León State Government (Table 2; SDS, 2016).From November 1992 to April 2003, and in accordance with EPA EQOA-0880-047, Thermo Environmental Instruments Inc. (TEI) model 49 UV photometric analysers were used to measure O 3 with a stated precision of less than ±2 ppb O 3 and a detection limit of 2 ppb O 3 .Similarly, in accordance with RFNA-1289-074, TEI model 42 chemiluminescence detectors were used to measure NO-NO 2 -NO x with a stated precision of less than ±0.5 ppb NO and a detection limit of 0.5 ppb NO.In May 2003, replacement TEI model 49C O 3 and model 42C NO-NO 2 -NO x analysers were operated as above, with a stated precision of better than ±1 ppb O 3 and ±0.4 ppb NO and detection limits of 1 ppb O 3 and 0.4 ppb NO.To rule out instrumentation influences on the air pollutant trends determined, long-term trends based on annual averages were compared with those derived using 3-year running averages, in accordance with Parrish et al. (2011) and Akimoto  (Radian, 2000), which is based on the US EPA AP-42 emissions factors categorisation (EPA, 1995).The emissions factors are regionalised for each Mexican state, based upon on-site measurements and survey information.Updates to the emissions factors have been conducted for each released NEI, although no changes in the methodology were implemented between the 1999 and 2008 base years.Overall, the mobile emissions were estimated using the MOBILE6-Mexico model (EPA, 2003).The emissions from point sources were derived using the annual operation reports submitted to SEMARNAT.The emissions from area sources were obtained using the categorisation of Mexican area sources and the regionalised AP-42 emissions factors.
The MCMA emissions inventories have been developed with a 2-year frequency since 1996 and were obtained from the MCMA Environment Secretariat website (http://www.aire.cdmx.gob.mx/).The methodology used to construct the MCMA inventory estimates is consistent with that used in the NEI (SEDEMA, 2016a), which is based on the AP-42 EPA emissions factors.However, more speciated emissions factors have been developed in each released version, considering updates in the local industrial activity, survey information, and field measurement campaigns.To date, the only significant change in the methodology is the replacement of the Mobile6-Mexico model with the MOVES model to obtain the 2014 base year mobile emissions (SEDEMA, 2016b).As for the MCMA inventories, more speciated emissions factors than those contained in the NEI were developed to produce the MMA emissions inventory 2013 base year (SDS, 2015); although, estimates of mobile emissions were obtained with the Mobile6-Mexico model (EPA, 2003).

Analytical methods
SIMA, SIMAT (Atmospheric Monitoring System of the MCMA), and SIMAJ (Atmospheric Monitoring System of the GMA) instrumentation recorded O 3 data every minute, which were then validated and archived as 1 h averages.Total SIMA O 3 data capture by year and site is shown in Fig. S3.Data capture averaged during 1993-2014 ranged from 82.6 % at GPE (see Table 2 for all site abbreviations) to 93.3 % at SNB, with data capture < 50 % during 1998-2000 at GPE, in 1998 at SNN, and in 1999 at OBI.A threshold of 75 % data capture was defined to consider data valid and representative (ProAire-MMA, 2008;Zellweger et al., 2009;Wilson et al., 2012).All data were processed with hourly averages used to determine daily averages, which were used to calculate monthly averages, from which yearly averages were obtained.

Data analysis methods
The SIMA, SIMAT, and SIMAJ O 3 data sets were analysed extensively using the openair package v1.1-4 (Carslaw and Ropkins, 2012) for R software v3.1.2(R Core Team, 2013).In this study, the openair functions windRose, timeVariation, and TheilSen were used to analyse air pollution data.Briefly, the windRose summarises wind speed and wind direction by a given timescale, with proportional paddles represent- ing the percentage of wind occurrence from a certain angle and speed range.The timeVariation function was used to obtain normalised daily cycles by season, and weekly cycles, with the 95 % confidence intervals in the cycles calculated from bootstrap resampling, which accounts for better estimations for non-normally distributed data (Carslaw, 2015).Finally, long-term trends in air pollutants in the MCMA, GMA, and MMA were computed with the TheilSen function, which is based on the non-parametric Theil-Sen method (Carslaw, 2015;and references therein).The Theil-Sen estimate of the slope is the median of all slopes calculated for a given n number of x, y pairs, while the regression parameters, confidence intervals, and statistical significance are determined through bootstrap resampling.It yields accurate confidence intervals despite the data distribution and heteroscedasticity and is also resistant to outliers.
The trends computed with openair were contrasted with those calculated using the MAKESENS 1.0 macro (Salmi et al., 2002) as follows.Firstly, the presence of a monotonic trend was tested with the non-parametric Mann-Kendall test.For the MCMA, GMA, and MMA, the available yearly data are n > 10; hence, positive values in the Z parameter correspond to positive trends and vice-versa for negative values of Z.The significance of the estimated trend was tested at α = 0.001, 0.01, 0.05, and 0.1 using a two-tailed test.Secondly, slopes of linear trends were calculated with the nonparametric Sen's method, which assumes linear trends, with a Q slope and a B intercept.To calculate Q, first the slopes of all data values were calculated in pairs, with the Sen's estimator slope as the median of all calculated slopes.Finally, 100(1 − α) % two-sided confidence intervals about the slope estimate were obtained based on a normal distribution.Comparisons of estimated trends from both approaches are shown in the Supplement Sect.S1.4 (Fig. S4).
The O 3 and other air pollutant time series were decomposed into trend, seasonal, and residual components using the seasonal trend decomposition technique (STL; Cleveland et al., 1990).STL consists of two recursive procedures: an inner loop nested inside an outer loop, assuming measurements of x i (independent) and y i (dependent) for i = 1 to n.The seasonal and trend components are updated once in each pass through the inner loop; each complete run of the inner loop consists of n (i) such passes.Each pass of the outer loop consists of the inner loop followed by a computation of the robustness weights, which are used in the following run of the inner loop to minimise the influence of transient and aberrant behaviour on the trend and seasonal components.The initial pass of the outer loop is performed with all robustness weights equal to 1, followed by n (0) passes of the outer loop.The Kalman smoother (KS) was used to provide minimum-variance unbiased linear estimations of observations and to impute missing data to satisfy the STL (Reinsel, 1997;Durbin and Koopman, 2012;Carslaw, 2015).Overall, statistical seasonal autoregressive and moving averages with annual seasonal components were employed.Statistical analyses were carried out with SPSS 19.0.
In order to carry out seasonal analyses of data, seasons were defined according to temperature records in the NH, as described previously (Hernández-Paniagua et al., 2015): winter (December-February), spring (March-May), summer (June-August), and autumn (September-November). Windsector analyses of data were performed by defining eight wind sectors each of 45 • starting from 0 ± 22.5 • .The lower bound of each sector was established by adding 0.5 • to avoid data duplicity.Data were assigned to a calm sector when wind speed was ≤ 0.36 km h −1 (0.1 m s −1 ).To assess regional transport, air mass back trajectories (AMBTs) were calculated using the HYSPLIT model v.4 (NOAA Air Resources Laboratory (ARL); Stein et al., 2015), with the Global NOAA NCEP/NCAR reanalysis data files on a latitude-longitude grid of 2.5 • , downloaded from the NOAA ARL website (http://ready.arl.noaa.gov/HYSPLIT.php).HYSPLIT frequency plots of 96 h AMBTs were constructed for every 6 h during the year 2014 with an arrival altitude of 100 m a.g.l.

Wind occurrence at the MMA
The MMA is highly influenced by easterly anticyclonic air masses that arrive from the Gulf of Mexico, especially during spring and summer (Fig. S5). Figure 2 shows the frequency count of 1 h averages of wind direction by site and season within the MMA during 1993-2014.At all sites, apart from OBI, the predominant wind direction is clearly E, which occurs between 35 and 58 % of the time depending on season.Easterly air masses are augmented by emissions from the industrial area E of the MMA, which are transported across the urban core and prevented from dispersing by the mountains located S-SW of the MMA.On average, the highest wind speeds are observed during summer at all sites.By contrast, calm winds of ≤ 0.36 km h −1 (0.1 m s −1 ) occurred less than 2 % of the time at all sites, most frequently in winter, and least frequently in summer.

Time series in O 3 and O x recorded within the MMA during 1993-2014
Within the MMA, the highest O 3 mixing ratios (1 h averages) are typically observed between April and September, whereas the lowest values are usually recorded between December and January (winter) (Fig. S6).Table S1 SNB and SNN, in 2001and 2002, respectively, to 46 ppb at OBI and STA in 2001and 2006, respectively.It is clear that the highest O 3 and O x mixing ratios were recorded when control of precursor emissions of VOCs and NO x were less stringent than subsequently.

Diurnal variations in O 3 and O x within the MMA
Here, O 3 diurnal variations were used to assess changes in the net O 3 production.Figure 3 shows daily profiles by season of O 3 , O x , NO, NO 2 , NO x , and SR averaged over the five sites within the MMA.O 3 generally dips during the morning rush hour due to titration with NO and mirrors the increase in NO 2 , which occurs around 07:00 in spring and summer and around 08:00 in autumn and winter.The 1 h difference in the O 3 dip derives from the change to daylight saving time during spring and summer.O 3 generally peaks during the enhanced photochemical period, around 13:00 in spring, 12:00 in summer (coincident with SR), and about 14:00 in autumn and winter.Similar profiles are observed for O 3 in all seasons, being negatively correlated with NO 2 (r = 0.93 (winter) to r = 0.97 (summer); p < 0.05), due to the rapid photolysis of NO 2 .Diurnal cycles of O x behave as O 3 , with the lowest values before the morning rush hour and the largest between midday (summer) and 15:00 (winter).During the daytime, O x and O 3 diurnal cycles are strongly correlated in all seasons, ranging from r = 0.97 in winter to r = 0.99 in autumn (p < 0.05), which suggests net O 3 production during the daytime.
O x amplitude values (AV d 's) derived from normalised daily cycles were used as a proxy to assess differences in the net O 3 production from site to site within the MMA.The normalised daily cycles were constructed by subtracting daily averages from hourly averages.Figure 4 shows normalised O x daily cycles.The lowest AV d 's in O x occur in winter consistent with reduced SR and low photolysis rates, with the largest values observed in summer.It is clear that during the year, the largest AV d 's are recorded at sites downwind of industrial emissions sources, in particular at STA, while the lowest AV d 's are observed at sites upwind.The larger AV d 's at downwind sites are interpreted to indicate higher net O 3 production, derived from the occurrence of photochemical processed air masses from the E sector.The AV d 's at upwind sites are less affected by emissions from the MMA and especially the industrial area.(Camalier et al., 2007;Zheng et al., 2007).Annual average cycles for those meteorological variables.O 3 and O x were constructed by averaging monthly averages for the same month during the studied period.Figure 5a shows that O 3 exhibits the maxima during spring and minima in winter, with a downward peak in early autumn, a behavioural characteristic of tropospheric O 3 in the NH.O x peaks in spring and dips in summer; although it is evident that NO x emissions lead to apparently similar O x   O 3 AV s of 19.0 ppb observed in 2014.Figure 5b shows that O 3 AV s decreased significantly at all sites between 1993 and 1997-1998, at rates from 0.78 ppb O 3 yr −1 at GPE to 2.28 ppb O 3 yr −1 at SNN (Fig. 5c).O 3 AV s 's have increased constantly (p < 0.05) at all sites since 1998, ranging from 0.90 ppb O 3 yr −1 at GPE to 0.75 ppb O 3 yr −1 at SNN. O x AV s s exhibited no discernible trends at all sites for the whole studied period, although SNN showed a significant (p < 0.05) decline during 1993-2001 (1.5 ppb yr −1 ) and STA showed an increase during 2004-2010 (1.3 ppb yr −1 ).The trends in O x follow those observed for NO x at SNN and STA during 1993-2014, which indicates that nearby indus-trial emissions have a significant contribution to the observed O x levels within the MMA.

Long-term trends in O 3 and O x within the MMA during 1993-2014
Quantifying the absolute changes in ground-level O 3 in response to trends in its precursor emissions is crucial to evaluate the impacts of air quality control (Parrish et al., 2009;Simon et al., 2015).The growing economy within the MMA has increased O 3 precursor emissions from point and area sources due to the limited emissions control programs (INEGI, 2015;SDS, 2015).Moreover, predominant E-SE winds throughout the year transport primary pollutants and their oxidised products downwind from the industrial area, which can offset reductions in emissions from other sources.
Here, to characterise changes in net O 3 production during 1993-2014 within the MMA in response to changes in its precursor emissions, long-term trends for daytime (06:00-18:00 CDT) O 3 and O x measurements were derived by averaging data in seasonal periods.Seasonal averaging was used to minimise variability inherent in longer-term averages and the de-seasonalisation process avoids confounding overall trends, especially when seasons exhibit opposite trends (Parrish et al., 2009).
Figure 6 shows seasonal trends in O 3 within the MMA, and Table 3 summarises the parameterisation of the trends.Significant increases (p < 0.1) in O 3 are observed at all sites, apart from STA, in spring and summer, while in autumn, O 3 increases significantly only at SNN and SNB.The increases in O 3 range from 0.26 ppb yr −1 in spring at OBI to 0.47 ppb yr −1 in summer at SNN.Overall, the lowest O 3 growth rates are observed at the urban background GPE site, whereas the largest ones are at the industrial SNN site.It is worth noting that only SNN and OBI exhibit significant increases in autumn, despite a decrease in the frequency of high wind speeds (> 20 km h −1 ).The existence of significant trends at all sites during spring-summer, except for OBI, is consistent with the downwind transport of industrial emissions and the high frequency of photochemical processed air masses with NE-E-SE origin, where the industrial area is located (Fig. S9).
Seasonal trends in O x are shown in Fig. 7, with the parameters of the trends listed in Table 3.Consistent with the seasonal O 3 trends observed, significant increases (p < 0.1) in O x within the MMA are determined in spring at all sites except for STA and range from 0.02 ppb yr −1 at OBI to 0.67 ppb yr −1 at SNB.It is worth noting that the industrial SNN and SNB sites show significant increases in O x in all seasons, with the lowest growth rates in winter and the largest in summer and spring.Moreover, STA exhibits the only significant decrease in O x of 0.63 ppb yr −1 during winter.As for O 3 , the O x increasing trends are consistent with the transport of primary emissions during the high occurrence of NE-E-SE air masses at WS > 10 km h −1 , which is highlighted during the photochemical season (April-September).Furthermore, the small shift in wind direction at STA to NW during winter coincides with the only observed decrease in net O 3 production within the MMA, which confirms that O 3 precursors are emitted E of the MMA.This also makes evident that increasing upwind industrial emissions have offset reductions in emissions from on-road sources as revealed by the decline in NO x evident at OBI.  (Stephens et al., 2008;Jaimes-Palomera et al., 2016).It should be noted that the peak value of O 3 for the GMA in winter and spring occurs an hour or so earlier than for the MMA and MCMA, which is consistent with higher VOC/NO x emissions ratios in the GMA (Kanda et al., 2016).As might be anticipated, larger AV d 's of 76.9±1.6 ppb O 3 are observed for the MCMA than for the GMA (46.1 ± 1.0 ppb O 3 ) and MMA (37.6 ± 0.4 ppb O 3 ), related to the levels of emissions of the O 3 precursors.

Comparison
The O x profiles show a trough during the morning rush hour and a peak between 12:00 and 14:00 in all urban areas.Despite large variations between weekday and weekend NO x mixing ratios in the three urban areas as shown in Fig. 8, no significant differences (p > 0.05) in O 3 and O x are observed in any of the metropolitan areas between O 3 and O x weekend and weekday AV d 's.Stephens et al. (2008) suggested that the most plausible explanation for the lack of weekend O 3 effect at MCMA during 1987-2007 is a simultaneous decrease in NO x and VOC emissions on weekends since the sole decrease in NO x emissions under VOC-limited conditions would lead to an increase in O 3 not observed.Similarly, a VOC-limited O 3 production regime was reported for the MMA by Sierra et al. (2013), whereas Kanda et al. (2016) reported that in the GMA the O 3 production lies in the region between VOC and NO x sensitivity.Therefore, it can be suggested that simultaneous decreases in NO x and VOC emissions on weekends in the GMA and MMA explain the similar behaviour in O 3 and O x in the MCMA.Moreover, a change in NO x -limited O 3 production regime on weekends in the three urban areas www.atmos-chem-phys.net/17/9163/2017/Atmos.Chem.Phys., 17, 9163-9185, 2017  The overall trends in net O 3 production during 1993-2014 in the MCMA and GMA are consistent with the significant (p < 0.05) annual decreases in NO x of 1.21 and 1.25 ppb yr −1 , respectively (Fig. 10).By contrast, while average NO x levels have increased annually in the MMA at the rate of 0.33 ppb yr −1 (p < 0.05), the average net O 3 production has remain steady.It is possible that the non-linear response in O x to the changes in NO x in an environment of high NO x mixing ratios (> 60 ppb) displaces the chemical equilibrium to favour NO as the dominant component of NO x , which does not account for the levels of O x (Clapp and Jenkin, 2001).It is also possible that the O x trends derived from the combined data set for the MMA do not represent local observed trends because there is a compensating effect between O x reductions and increases.

Compliance with the 1 and 8 h Mexican standards
for O 3 within the MMA Between 1993 and 2014, there were two official standards for maximum permitted mixing ratios of O 3 in Mexico: (i) a running 8 h average of 80 ppb, not to be exceeded more than four times per calendar year, and (ii) a 1 h average of 110 ppb (NOM-020-SSA1-1993).Since 19 October 2014, the maximum permitted O 3 levels were lowered to a running 8 h average of 70 ppb and a 1 h average of 95 ppb, (NOM-020-SSA1-2014).However, because both standards are applicable for whole calendar years, the old permitted O 3 levels were used in this study to determine the number of annual exceedances of both O 3 standards.Figure 11 shows that within the MMA, the O 3 1 h average and the running 8 h standards were frequently exceeded (INE, 2011;SEMARNAT, 2015).The largest number of exceedances occurs at STA, followed by SNB, GPE, and OBI, whereas the fewest breaches have been observed at SNN markedly since 2004.However, there have been three periods of clear decreased exceedances at all sites (except STA in 2014(except STA in ) during 1994(except STA in -1995(except STA in , 1999(except STA in -2000(except STA in , and 2012(except STA in -2013, which are consistent with marked changes in the national GDP during economic recessions in Mexico (Fig. S10a).However, national GDP exhibits a notable decrease during the 2008-2009 global economic recession; only in 2009 do the O 3 annual exceedances within the MMA seem to follow (Fig. S10b).Therefore, if O 3 levels continue to increase within the MMA, as determined in the long-term trend assessment, an increase in peak O 3 mixing ratios is also likely to occur.Hence, to analyse changes in peak O 3 , daily maxima 1 h averages from 1993 to 2014 were used to determine seasonal trends in peak levels.Figure 12 shows trends in 1 h daily maxima and Table 5 lists the parameters of the trends.Daily maxima O 3 1 h averages increased significantly (p < 0.05) in spring and summer at all sites, except for STA, and also in autumn at the industrial sites SNN and SNB.The largest increases in the daily maxima are seen at SNN, where similar increases between 0.85 and 0.93 ppb yr −1 are determined between spring and autumn.SNB exhibits slightly lower growth rates in spring and summer but a large difference in autumn.We have shown that predominantly E-SE winds transport photochemically processed air masses to SNN and SNB during spring-summer leading to the observed exceedances.Moreover, the change in the wind occurrence in autumn at SNB leads to a lower growth rate than at SNN, where the calmest winds during the whole year drive the largest increase, interpreted to be due to the photochemical processing of precursors emitted locally.The GPE and OBI sites exhibit increases only in spring and summer, with the lowest increases of all sites of 0.48 ppb yr −1 determined at OBI in spring, which contrasts with the largest increase at OBI during the same season.However, such increases are consistent with an increase in the occurrence of NE and E air masses at high speeds (> 10 km h −1 ) during spring-summer.STA shows a significant decrease in the maximum daily O 3 1 h averages of 0.35 ppb yr −1 in winter, which is consistent with an increase in the occurrence of NW air masses at WS < 5 km h −1 , loaded with high NO x mixing ratios (50 ppb) that promote the O 3 titration.(Carslaw, 2015).

Strategies for air quality control in Mexico
The Mexican environmental authorities have focused largely on improving the air quality within the MCMA since 1986 by implementing numerous strategies to control primary emissions, but they have paid less attention to other large metropolitan areas in Mexico (PICCA, 1990; ProAire-MCMA, 2011).Control measures have been designed based on NAEI and local emissions inventory data, which possess significant uncertainties (Arriaga-Colina et al., 2004;Velasco et al., 2007;Kanda et al., 2016).However, despite these uncertainties, the emissions control strategies have helped to reduce O 3 levels within the MCMA since 1991-1992(ProAire-MCMA, 2001).Here, we describe the most effective measures introduced to control O 3 precursor emissions within the MCMA, and then we discuss potential benefits of implementing such measures within the MMA.
From 1993 to 2014, NO x levels within the MCMA decreased at a rate of around 1.2 ppb yr −1 (1.6 % yr −1 ) as determined from ground-based measurements.This decline is remarkably consistent with the decrease in the NO 2 column over the MCMA of 1.6 % yr −1 during 2005-2014 reported by Duncan et al. (2016).The decrease in NO x was driven largely by reductions in emissions from on-road sources, in response to the introduction of mandatory three-way catalytic converters in new vehicles in 1993 (NOM-042; SE-MARNAT, 1993b) and by the introduction of a no-driving day and more stringent exhaust emissions inspection programs for private cars in 1989 (NOM-041; SEMARNAT, 1993a).The NO x reduction measures also required public transport vehicles to switch from engines fuelled by petrol to engines fuelled by LP gas.New road corridors were designed for improving the intracity transport and the public transport fleet was renewed (ProAire-MCMA, 2001).For industrial sources, the switch from fuel oil to LP gas fuel, relocation of highly polluting industries away from the MCMA, and im-    plementation of regular inspection programs of NO x emission for industrial and area sources were also implemented (ProAire-MCMA, 2001).While the outlook for NO x levels within the MCMA is clear, studies of VOC levels have reported no conclusive trends.For instance, Arriaga-Colina et al. (2004) reported a decrease in VOCs of around 10 % from 1992 to 2001 over the N MCMA, while Garzón et al. (2015) reported that on average VOCs increased over most of the MCMA between 1992 and 2002 but decreased by 2.4 ppb yr −1 between 2002 and 2012.However, the decrease in VOCs from 2002 to 2012 reported by Garzón et al. (2015) is consistent with a reduction in light alkane and aromatic levels during the morning rush hour reported by Jaimes-Palomera et al. (2016).Continuous measurements of VOCs have been recently introduced by the MCMA government, which precludes an assessment of VOC long-term trends.The measures implemented to control VOC emissions from on-road sources have included the reformulation of petrol with the reduction of highly reactive VOCs and addition of oxygenated compounds and fitting of three-way catalytic converters in all new vehicles (NOM-042; SEMARNAT, 1993b;ProAire-MCMA, 2001).For area sources, control measures include the introduction of vapour emissions control systems at petrol stations and the introduction of a LP gas leak detection program for the distribution network (ProAire-MCMA, 2011).As for NO x , industrial VOC emissions sources have been subject to regular emissions inspections and relocation of the most significant emitters (ProAire-MCMA, 2011).
Therefore, the moderate success on controlling O 3 levels within the MMA can be interpreted as due to the implementation of effective control measures on VOCs and NO x emissions.Thus, a comparison between VOCs and NO x trends derived from the NAEI and local emissions inventories with those determined from ground-level measurements can provide insight into further improvements in decreasing O 3 levels not only within the MCMA but also in other large metropolitan areas in Mexico.Within the MCMA, the NAEI NO x emissions trends are consistent with the decrease determined from ground-based measurements made by SIMAT, but the MCMA local inventory trends disagree with the SIMAT trends (Figs.S1 and 10).For VOCs, the NAEI and the MCMA inventories oppose measured trends in VOCs during 1993-2001(Arriaga-Colina et al., 2004;Garzón et al., 2015).This can be explained by underestimates of VOC emissions within the MCMA of a factor of 2-3 (Arriaga-Colina et al., 2004;Velasco et al., 2007).Such discrepancies suggest that significant improvements in NO x and VOC emissions inventories are still required to better inform O 3 control strategies.2006, 2011, 2014;SDS, 2015).This is underlined at OBI with the highest values of O x where the predominant wind direction is NE, consistent with the transport of emissions from the industrial area located to the NE and photochemical processing of air masses (Carrillo-Torres et al., 2017).The daily cycles of O 3 determined within the MMA are consistent with those reported for Los Angeles (VanCuren, 2015) and Toronto (Pugliese et al., 2014).In Toronto, the O 3 maxima were enhanced by the arrival of photochemically processed air masses transported from polluted wind sectors, and decreased during clear air masses.This behaviour is similar to that observed within the MCMA, with enhanced O 3 maxima during the occurrence of E-SE (polluted) air masses and decreased levels when W-SW (relatively clean) air masses occurred.

Origin of the O 3 annual cycles within the MMA
The annual O 3 cycles within the MCMA are consistent with the spring maxima and winter minima characteristic of the southeastern US regions (Strode et al., 2015) and follow the O 3 cyclic pattern at NH mid-latitudes (Monks, 2000;Vingarzan, 2004).However, they are different from O 3 annual cycles reported for the US west coast regions, particularly in California, where the maxima in the cycle occurs between June and August, driven by the local influence of precursor emissions on O 3 production and photochemical conditions (Vingarzan, 2004;Strode et al., 2015).The recurrent downward spikes in the O 3 annual cycles within the MMA between July and August result from high wind speeds (> 10 km h −1 on average) that disperse O 3 precursors and increase the boundary layer height (ProAire-MMA, 2008).The peak in O 3 observed in September is characteristic of humid regions and can be ascribed to an increase in OH radicals derived from the increment in RH during the rainy season (Lee et al., 2014).A marked increase in RH within the MMA during September is consistent with the increase in O 3 observed as reported by Lee et al. (2014).Over the Midwestern and eastern US regions, that O 3 peak has become less noticeable since 2000 (Zheng et al., 2007).The annual variability in O 3 within the MMA is strongly coupled to the economic conditions (GDP) in Mexico.For instance, the economic crisis of 1994-1996 caused a marked reduction in industrial emissions of VOCs and NO x , which is confirmed by the significant decrease in O 3 annual variations at all sites within the MMA (Tiwari et al., 2014;INEGI, 2016).During the global economic recession of 2008-2009, Castellanos and Boersma (2012) reported a reduction of 10-30 % in tropospheric NO 2 over large urban European areas, which is consistent with a faster decline of 8 ± 5 % yr −1 in the NO 2 column density during the same period for US urban regions (Russell et al., 2012).Increases in the NO 2 column density over the MMA as reported by Duncan et al. (2016) are explained by the gradual recovery of the economy since 1997 in Mexico.Moreover, increases in O 3 precursor emissions and in annual variability observed within the MMA are consistent with such economic growth.This clearly ex- plains the opposite trends in O 3 annual variations before and after the economic crisis within the MMA, with the lowest changes seen at the urban GPE site and the greatest ones detected for the SNN industrial site.

Increasing O 3 and O x levels within the MMA
Ground-based measurements made during 1993-2014 reveal significant (p < 0.05) increases in NO x within the MMA at all sites, apart from OBI, which exhibits a significant decrease (Fig. 13).Overall, the NO x increase within the MMA of 1.24 % yr −1 (0.33 ppb yr  2016).The influence of industrial emissions upon O 3 at the MMA becomes evident from the lowest NO x growth rate of 0.19 ppb yr −1 observed at GPE since OBI has few occurrences of air masses transporting pollutants from the largely industrialised areas throughout the year (Fig. 2).By contrast, the NO x decrease of −0.40 ppb yr −1 at OBI arises from decreases in emissions from on-road sources (SDS, 2015).The large growth rates in O 3 and NO x at SNN and SNB are explained by increasing emissions of O 3 precursors from a growing number of industries and the urban development E of the MMA.The most likely explanation for the O 3 in-crease at OBI is a reduced titration effect from decreasing NO x levels in combination with the non-linear response in O 3 production to decreasing NO x emissions under the VOCsensitive MMA airshed (Sierra et al., 2013;Menchaca-Torre et al., 2015).
The O x long-term trends during 1993-2014 within the MMA were consistent with those for O 3 at all sites.Decreases in NO x and O 3 observed between 1994 and 1996 were the response to the economic crisis during the same period in Mexico, when the GDP decreased by 5.9 %, providing additional evidence of the dominant role of industries within the MMA.Consistent with economic indicators, annual averaged petrol sales in the Nuevo León State in 1995 decreased by 2.4 % in relation to 1994, but increased linearly from 1996 to 2008 at an approximate rate of 98 800 m 3 petrol yr −1 (r = 0.90) (Fig. S11) (SENER, 2015).As for petrol sales, registered vehicles in Nuevo León show significant variations between 1993 and 1996 but have increased linearly since 1997 at a rate of around 100 000 vehicles yr −1 (r = 0.99).This confirms that despite the annual growth in the vehicular fleet, the fitting of three-way catalyst technology and the reformulation of petrol introduced in 1997 have controlled on-road primary emissions (ProAire-MCMA, 2001) The decreases in NO x observed at OBI and at all sites during the occurrence of W-SW-NW air masses reflect that if applied, stricter emissions controls such as those for on-road sources can lead to a significant abatement in primary emissions.It is clear that the industrial sources within the MMA must be subject to similar emissions control measures as those implemented within the MCMA for effectively reducing the O 3 levels.

The opposite O 3 trends in Mexican urban areas
The comparison of O 3 and O x trends in MMA, GMA, and MCMA reveals different emissions trends at each of the cities studied.The trends in O 3 reported in this study for the MCMA agree with the reduction of 20 ppb O 3 during 1991-2011 for the MCMA (Jaimes et al., 2012), and with the reduction of 8 ppb O 3 during 2000-2011 for the MMA (Benítez-García et al., 2014).In the GMA, the no-trend status in O 3 determined here is in contrast with the increase of 12 ppb O 3 during 2000-2011(Benítez-García et al., 2014), which is due to the different periods assessed in both studies.Decreases in O 3 in urban US areas arise from effective control of O 3 precursor emissions (Strode et al., 2015), which has occurred at the MCMA.
Figure 10 shows that NO x decreased significantly within the MCMA (1.57% yr −1 ) and the GMA (1.83 % yr −1 ) during 1993-2014 and 1996-2014, respectively, but increased within the MMA (1.83 % yr −1 ) during 1993-2014.Such NO x trends are within the range of the trends in the NO 2 column density reported by Duncan et al. (2016) in Table S9, which reveals an increase of 0.78 ± 1.12 % yr −1 for the MMA, but decreases of 1.82 ± 0.84 % yr −1 for the GMA and of 0.10 ± 1.67 % yr −1 for the MCMA, all during 2005-2014.To date, long-term trends in VOCs have only been reported in the MCMA, with an average decrease of ca.2.4 ppb yr −1 since 2002, mostly in propane, ethanol, and acetone (Garzón et al., 2016), while there are no studies of longterm trends in VOCs within the MMA and the GMA.
It has been shown that O 3 and O x decreases within the MCMA have been driven by reductions in NO x and VOC emissions and that the implemented strategies described in Sect.4.1 have proved to be effective in controlling primary emissions (ProAire-MCMA, 2011;Jaimes-Palomera et al., 2016).By contrast, growing industrial emissions within the MMA must be subject to stringent controls to abate O 3 levels.In the GMA, where the industrial activity is lower than in the MCMA and MMA (Kanda et al., 2016), the policies introduced on a national scale for controlling on-road source emissions have resulted in the decrease in NO x emissions and in the stabilisation of O 3 levels.The results presented here demonstrate the merits of the assessment and analysis of long-term O 3 levels, which can be used by environmental authorities to revise and to redesign programs and policies to improve air quality.Continuing with ground-based O 3 and NO x monitoring is strongly recommended to better understand the response of changes in primary emissions to further changes in local and regional O 3 levels.Monitoring of VOCs in the GMA and MMA is also recommended since the VOC emissions data reported in the NAEI possess significant uncertainties.Finally, in accordance with the results presented here, we recommend preferentially reducing VOC emissions, which may limit O 3 production in response to a decrease in the VOC / NO x ratio.However, simultaneously reducing NO x will have added health benefits of less NO 2 .

Conclusions
Diurnal and annual cycles and long-term trends in O 3 and O x within the MMA are interpreted as a response to changes in NO x and VOCs emissions, photochemistry, and meteorology.Continuous high-frequency and high-precision O 3 and NO x data recorded during 1993-2014 at five sites within the MMA and at 29 sites within the MCMA and during 1996-2014 at 10 sites within the GMA were used to calculate longterm trends.Within the MMA, the greatest mixing ratios in O 3 were recorded during E and SE winds, at sites downwind of significant precursor emissions from industrial sources.By contrast, the lowest O 3 mixing ratios were recorded at SNN, and for all sites were observed for the W and SW sectors, where air masses travel from central Mexico over 100-300 km of semi-arid and sparsely populated area.Maximum daily 1 h values of O 3 and O x increased significantly at GPE, SNN, and SNB, owing to increasing emissions of precursors, while at OBI increasing O 3 and decreasing O x trends arise from the non-linear response to decreasing NO x emissions from on-road sources.

a
Average period.b Not to be exceeded more than four times in a calendar year.c Running average.

Figure 1 .
Figure 1.(a) The MMA, MCMA, and GMA in the national context.(b) Topography of the MMA and distribution of the five monitoring sites over the area.(c) The five monitoring sites in relation to primary and secondary motorways, industries, and major residential areas.The red arrows show the predominant wind direction at each site from 1993 to 2014.

Figure 2 .
Figure 2. Frequency of counts of measured wind direction occurrence by season and site within the MMA during 1993-2014.

3. 4
Annual cycles of O 3 and O x within the MMA Annual variations in O 3 and O x are correlated positively with the seasonality of temperature, RH, and SR

Figure 3 .
Figure 3. Seasonal average daily profiles for O 3 , O x , NO x , NO, NO 2 , and SR within the MMA during 1993-2014.The shading shows the 95 % confidence intervals of the average.

Figure 4 .
Figure 4. Seasonal O x de-trended daily profiles within the MMA during 1993-2014.De-trended O x daily cycles were constructed by subtracting daily averages from hourly averages to remove the impact of long-term trends.

Figure 5 .
Figure 5. (a) Annual cycles of O 3 , temperature, rainfall, RH, and SR constructed by averaging records from 1993 to 2014 for a 1year period.(b).Trends in AV s of O 3 recorded at the five monitoring sites within the MMA from 1993 to 2014.The decline in AV s observed is due to the economic crisis experienced in Mexico during 1994-1996, followed by persistent increases in AV s since 1998.(c) Annual rates of change in O 3 AV s by site before and after the 1994-1996 economic crisis.

Figure 6 .
Figure 6.Seasonal trends in O 3 within the MMA during 1993-2014.Each data point represents the average of the 3-month period that defines the season.The continuous lines show the Sen trend.

Figure 7 .
Figure 7. Seasonal trends in O x within the MMA during 1993-2014.Each data point represents the average of the 3-month period that defines the season.The continuous lines show the Sen trend.

Figure 8 .
Figure 8. Seasonal average diurnal cycles of O 3 , O x , and NO x during 1993-2014 for the MCMA and the MMA, and between 1996 and 2014 for the GMA.The shading shows the 95 % confidence intervals of the average, calculated through bootstrap resampling(Carslaw, 2015).

Figure 9 .
Figure 9. Seasonal trends in O 3 and O x for the MCMA and MMA during 1993-2014 and for the GMA during 1996-2014.Each data point represents the average of the 3-month period that defines the season.The dashed lines show the Sen trend.

Figure 10 .
Figure 10.Trends in NO x in the MCMA and MMA during 1993-2014 and in the GMA during 1996-2014.The dashed lines represent the Sen slopes.All trends are statistically significant at p < 0.05.

Figure 11 .
Figure 11.Annual exceedances of the O 3 NOM for 1 h averages (110 ppb) and 8 h running averages (80 ppb) at the five monitoring sites within the MMA from 1993 to 2014.

Figure 12 .
Figure 12.Seasonal trends in 1 h O 3 daily maxima at the MMA during 1993-2014.Each data point represents the average of the 3-month period that defines the season.The dashed lines show the Sen trend.

Figure 13 .
Figure 13.Long-term trends for NO x at the five monitoring sites within the MMA during 1993-2014.The dashed lines represent the Sen slopes.Annual NO x rates of change are expressed as the slope (m) of the dashed lines in units of ppb yr −1 .Levels of confidence are represented as + = p < 0.1, * = p < 0.05, * * = p < 0.001, and * * * = p < 0.001.

Table 1 .
Air quality limit values stated in Mexican legislation.

Table 2 .
Site description and location of the five monitoring sites within the MMA.
Reaction with O 3 rapidly converts NO to NO 2 , and therefore mixing ratios of odd oxygen (O x = O 3 +NO 2 ) were calculated to account for O 3 stored as NO 2 for each hour during 1993-2014 at the five sites within the MMA (Table S2; Fig. S7).Minimum values of O x ranged from 2 ppb, observed at all sites during 1993-2014 to 13 ppb at OBI in 2007.Maxi-mum values of O x ranged from 99 ppb at SNN in 2002 to 330 at OBI in 1993.O x annual averages varied from 23 ± 17 ppb at SNN in 2002 to 51±27 ppb at OBI and at STA in 2001 and 2006, respectively, whereas O x annual medians ranged from 21 ppb at of MMA O 3 and O x weekly profiles with those in MCMA and GMA O 3 production varies from city to city in response to local NO x and VOC emissions.Assessment of weekly profiles of O 3 and O x may provide insights into the geographic response in net O 3 production to diurnal variations in precursor emissions.Hourly O 3 and O x averages were used to construct weekday and weekend average profiles for the MCMA from 1993 to 2014, and for the GMA from 1996 to 2014. Figure 8 compares weekly O 3 and O x profiles by season within the MMA with those for the MCMA and GMA.In each case, and consistent with observations in other major urban areas of NA, the lowest O 3 mixing ratios occur during the morning rush hour due to O 3 titration with NO emitted from on-road sources, whereas peak values of O 3 are apparent after midday during periods of enhanced SR

Table 3 .
Results for O 3 and O x long-term trends expressed in ppb yr −1 for 1993-2014 at the five sites within the MMA by season.
a Level of significance p < 0.1.b Level of significance p < 0.05.c Level of significance p < 0.001.d Level of significance p < 0.001.

Table 4 .
Results for O 3 and O x long-term trends by season expressed in ppb yr −1 during 1993-2014 for the MCMA and MMA and during 1996-2014 for the GMA.Level of significance p < 0.1.b Level of significance p < 0.05.c Level of significance p < 0.001.d Level of significance p < 0.001. a -level O 3 and O x variations within the MMA The O 3 and O x diurnal variations result from the particular chemical environment and meteorological conditions at each monitoring site within the MMA.Thus, the largest O 3 and O x mixing ratios, except for OBI, are typically observed for air masses from the E and SE wind sectors, whereas at www.atmos-chem-phys.net/17/9163/2017/Atmos.Chem.Phys., 17, 9163-9185, 2017

Table 5 .
Results for O 3 daily maximum long-term trends by season in ppb yr −1 during 1993-2014 at the five sites within the MMA.
a Level of significance p < 0.1.b Level of significance p < 0.05.c Level of significance p < 0.001.d Level of significance p < 0.001.
Duncan et al. (2016)4 is larger than the increase in the NO 2 column density over the MMA of around 0.78 % yr −1 during 2005-2014 reported byDuncan et al. (2016); although both have indicated a significant increase in the NO x levels at least since 2005.The largest increases in NO x correspond to industrial sites SNN (0.51 ppb yr −1 ) and SNB (0.74 ppb yr −1 ).This is interpreted as a response to growing industrial activity, in combination with flexible emissions regulations within the MMA (INEGI,