Weekly patterns of México City ’ s surface concentrations of CO , NO x , PM 10 and O 3 during 1986 – 2007

Surface pollutant concentrations in M éxico City show a distinct pattern of weekly variations similar to that observed in many other cities of the world. Measurements of the concentrations of carbon monoxide (CO), nitrogen oxides (NOx=NO+NO2), particulate matter smaller than 10 μm (PM10), and ozone (O3) collected hourly over 22 years (1986–2007) at 39 urban monitoring locations were analyzed. Morning concentrations of CO, NO x, and PM10 are lower on Saturdays and even more so on Sundays, compared to workdays (Monday–Friday), while afternoon O 3 concentrations change minimally and are occasionally even higher. This weekend effect is empirical evidence that photochemical O3 production is NOx-inhibited, and to the extent that emissions of CO are proportional to those of reactive volatile organic compounds (VOCs), it is VOC-limited, at least in the urban areas for which the monitoring stations are representative. The VOC-limitation has increased in the past decade, due to decreases in the concentrations of CO (and presumably VOCs) and consequent decreases in the CO/NO x and VOC/NOx ratios. Enhancements of photolysis frequencies resulting from smaller weekend aerosol burdens are not negligible, but fall short of being an alternate explanation for the observed weekend effect. The strength of the weekend effect indicates that local radical termination occurs primarily via formation of nitric acid and other NO x-related compounds, some of which (e.g. peroxy acyl nitrates) can contribute to the regional NOx budget. While VOC emission reductions would be most effective in reducing local O 3 production, NOx emission reduction may be more important for controlling regional oxidants. Correspondence to: S. Madronich (sasha@ucar.edu)


Introduction
The atmosphere of México City has received considerable scientific attention in recent years, foremost because of concerns about the potential health effects of air pollutants on its ∼20 million inhabitants, and also because it may be to some extent representative of current and future conditions in other megacities undergoing rapid economic development.The city's tropical high altitude location (19 • N, 2.2 km above sea level) is conducive to fast photochemistry forming secondary pollutants such as ozone (O 3 ) and particulate matter (PM).Several intensive measurement campaigns have characterized the main aspects of the meteorology and chemical composition, including MARI (LANL/IMP, 1994), IMADA/AVER (Doran et al., 1998), MCMA-2003(Molina et al., 2007), and in 2006 MILAGRO (Molina et al., in preparation, 2008).An air quality monitoring network was established in 1986, and has helped document long-term reductions of some pollutants following the institution of various emission-reduction programs (INE, 1998).
One of the issues most relevant to the design of emission reduction policies for urban areas is whether the formation of O 3 is more sensitive to emissions of nitrogen oxides (NO x ) or volatile organic compounds (VOCs).It is well known (e.g.Finlayson-Pitts and Pitts, 1986) that O 3 formation depends non-linearly on these emissions, and is maximal when VOC/NO x molar ratios are in the range of 5-15, the exact value depending on various conditions.At higher VOC/NO x ratios, O 3 production is limited by, and therefore sensitive to, the available NO x .At lower ratios it is limited by VOCs and, at sufficiently high NO x , even inhibited by any additional NO x (due to the reactions NO+O 3 →NO 2 +O 2 and OH+NO 2 →HNO 3 ).However, O 3 formation is also sensitive to other factors such as detailed VOC speciation and environmental conditions, so the direct measurement of VOC/NO x ratios is insufficient to establish whether the chemical regime is VOC-or NO x -limited.Sillman (1995) proposed using Published by Copernicus Publications on behalf of the European Geosciences Union.
S. Stephens et al.: Mexico City pollution weekend effect several other concentration ratios as indicators of NO x or VOC sensitivity involving, in addition to O 3 , total reactive nitrogen (NO y ) as well as photochemically produced formaldehyde (CH 2 O), nitric acid (HNO 3 ), and hydrogen peroxide (H 2 O 2 ).The VOC or NO x sensitivities can also be assessed from chemistry-transport modeling simulations, in which the emissions of VOCs and NO x are varied around central estimates and the response of O 3 concentrations is examined.
For México City the VOC or NO x sensitivity has been a subject of considerable debate.Molina et al. (2002a) reviewed ambient VOC/NO x measurements, emission inventories, and smog chamber experiments with captive air, to conclude that O 3 formation appeared to be NO x sensitive.They however recognized that this evidence is not conclusive because of the relative scarcity of VOC/NO x measurements and over-simplifications inherent in representing the full ozone photochemistry by this simple ratio; they also noted that VOC concentrations have been decreasing more rapidly than those of NO x over the previous decade, so the chemical regime may have been changing.Measurements of several sensitivity indicators (O 3 , NO x , and NO y ) were made 14-25 April 2004 at a location within México City (Santa Ursula) by Torres-Jardón (2004), with their ratios indicating VOC-sensitive conditions.During MILAGRO, Nunnermacker et al. (2008) measured H 2 O 2 concentrations typically below 1 ppb in urban overflights by the G-1 aircraft and at the T1 suburban supersite, consistent with a VOC-sensitive, high NO x regime.One modeling study (West et al., 2004, see their Fig. 6) found that increasing VOC emissions by as much as a factor of four resulted in O 3 increases of only 12-25%, while more recent modeling studies (Lei et al., 2007(Lei et al., , 2008;;Tie et al., 2007) found much stronger VOC sensitivity.
A separate empirical assessment of the response of O 3 to emission changes can be obtained by the weekend-workday differences in the emissions of O 3 precursors, and the resulting differences in O 3 concentrations.Generally, emissions of NO x and VOCs are lower on weekends, while in many locations (though not all) the weekend O 3 concentrations are minimally lower, or even higher, than on workdays.Observations of this effect have been made at many locations throughout the world, e.g. for the US in New York and New Jersey (Cleveland et al., 1974;Bruntz et al., 1974), the Baltimore-Washington area (Lebron, 1975;Jacobson, 1975), Southern California (Blanchard and Tanenbaum, 2003;Qin et al., 2003;Fujita et al., 2003;Chinkin et al., 2003), Central California (Blanchard and Fairley, 2001;Marr and Harley, 2002;Murphy et al., 2007), Northern California (Altshuler et al., 1996), Atlanta, Chicago, and Philadelphia (Pun et al., 2003), andPhoenix (Atkinson-Palombo et al., 2006); in Canada near Vancouver (Pryor and Steyn, 1995) and Toronto (Beaney and Gough, 2002), in Chile (Jorquera et al., 2000), Switzerland (Brönnimann and Neu, 1997), France (Pont and Fontan, 2001), the UK (Jenkin et al., 2002), Greece (Riga-Karandinos et al., 2006), India (Debaje and Kakade, 2006), and Nepal (Pudasainee et al., 2006).A weekend effect in the NO 2 column amount has also been detected by a satellitebased instrument over urban and industrial regions of the US, Europe, and Japan (Beirle et al., 2003).For México City, Muñoz et al. (2007) have shown statistically significant variations in O 3 concentrations as a function of day of the week for the years 1990-2006. Torres-Jardón (2004) found that at the Santa Ursula station during 1-30 April 2004 weekend NO concentrations (6-9 a.m.) were 37% lower than on workdays, while maximum O 3 concentrations were only 10% lower, consistent with a VOC-limited regime.
Here, we examine for México City the weekly patterns over 22 years  of NO x , carbon monoxide (CO, as a proxy for VOCs), O 3 and PM 10 (PM smaller than 10 µm, since 1993) concentrations analyzed from surface measurements at 39 urban locations (see Sect. 2).The differences between workdays (Monday-Friday) and weekends (Saturday and Sunday) are shown in Sect.3, while Sect. 4 discusses possible reasons for these patterns in terms of our understanding of the prevailing photochemical regime.The implications for urban and regional air quality are discussed in Sect. 5.

Data availability
Continuous monitoring of air pollutants in México City began in 1986 with the establishment of several networks (INE, 1998), now numbering 39 stations, to measure surface concentrations of O 3 , NO x , NO 2 , CO, SO 2 , TSP, and PM 10 , and surface meteorology.Hourly data are archived by the Government of México City (SIMAT, 2007).The performance of the air quality monitoring network has been reviewed periodically by the US Environmental Protection Agency, and a recent report concluded that the monitoring system is overall accurate and well implemented (GDF, 2004).A data screening procedure was implemented to eliminate possible values falling far outside realistic bounds.For NO x , NO 2 , and O 3 , allowed values were between 2 ppb and 1 ppm, for CO between 10 ppb and 100 ppm, and for PM 10 between 0.1 and 1000 µg/m 3 .These wide ranges should not be construed as actual data ranges, but rather are merely additional steps to screen out possible artifacts.
The large record of surface measurements allows the analysis and interpretation of temporal patterns on many time scales, including daily, weekly, seasonal, and long-term variations.Some averaging was carried out to reduce the effects of temporal and spatial variability and thus to bring out the more persistent temporal patterns, as follows.Values from individual stations were averaged together by five city sectors (see SIMAT, 2007 ically 1986-1992, 1993-2000, and 2001-2007.The availability of the data is summarized in Table 1, which gives the number of valid days summed over the stations of each sector, for each year.The data record is clearly more complete in the recent years, particularly for PM 10 and NO x .In all cases, relative changes (percents) were calculated as the deviations between average absolute values, rather than as the average of relative changes between individual values.For example, the average difference (%) between Sunday and Wednesday O 3 values in 2007 was computed by calculating the 2007 average Wednesday O 3 , then the 2007 average Sunday O 3 , and finally computing the percent difference between them (as opposed to computing the percent difference between each Wednesday and the previous or following Sunday, and then averaging the percent differences over the entire year).This procedure reduces the influence of short-term fluctuations in the weekend effect.Weekly patterns were also analyzed by Fourier multiple regression with nine fitting coefficients (average plus sines and cosines with periods of 7, 7/2, 7/3, and 7/4 days).This yielded the amplitude (positive or negative) of the weekly pattern, and its relative size (percent) compared to the average.Standard deviations (1σ where shown) were estimated using bootstrap resampling with replacement (Efron and Tibshirani, 1993).

Analytic derivation of fractional radical loss from NO x chemistry
The weekend-workday differences in the concentrations of CO, NO x , O 3 , and PM 10 can be used to estimate the fraction of the radicals lost via NO x chemistry, which in turn provides an indication of whether the O 3 production is VOC or NO x sensitive.The derivation of this fraction and the underlying approximations are presented here.Kleinman (2005) has shown that the instantaneous O 3 production rate, P O3 , is related to instantaneous NO x and reactivity weighted hydrocarbon (or VOC) concentrations and the radical production rate Q by: where L N is the radical loss due to NO x chemistry (e.g.OH+NO 2 →HNO 3 , and reactions of organic peroxy radicals with NO to form organic nitrates) rather than other processes (e.g.formation of peroxides at low NO x ).Because radical lifetimes are short, the radical production rate Q is essentially equal to the total radical loss, so that the ratio L N /Q is the fraction of the radical loss that occurs via NO x chemistry, with values larger than 0.5 for VOC-limited conditions, and smaller than 0.5 for NO x -limited conditions.It should be noted that Eqs.(1-3) more strictly represent the production of total odd oxygen, O x =O 3 +NO 2 (Sillman, 2008;Kleinman, L., priv. comm., 2008).These equations can be used to predict the response of O 3 or O x to changes in NO x and VOC emissions.With the simplified notation the change in O x concentration can be expanded as: where in the last equation for the sensitivities to NO x , VOCs, and Q (Eqs.1-3) were used.This equation can be solved for L N /Q: The terms on the right hand side can be estimated from the weekend effect with some additional approximations.First, we assume that the weekend effect for the instantaneous O x production, δP O x , is reflected to first order in the build-up of the afternoon O x concentrations considered here, so that δP O x ∼δO x .Second, as further discussed below, we assume that the weekend effect for VOC reactivity is similar to that for CO, δVOC∼δCO.Third, we assume that the change in the radical production rate is due mostly to changes in photolysis frequencies, so that δQ∼δJ .On this last point, we note that δQ also depends on the availability of photo-labile species, such as O 3 , CH 2 O, and HONO, which however are not likely to be larger on weekends, so δJ is probably an upper limit to δQ.With these approximations, Eq. ( 6) can be rewritten as: where all the terms on the right hand side can be measured or at least estimated from the weekend effect.

Results
The diurnal cycles of CO, NO x , O 3 , and PM 10 surface concentrations are shown in Fig. 1, averaged for all stations and all days over 2001-2007.For CO, NO x , and PM 10 the maximum values occur during the morning rush hours, followed by a decrease in the late morning due to lower emissions and the rapid growth of the planetary boundary layer (PBL) as re-cently reviewed by Shaw et al. (2007), a secondary maximum from the evening rush hours, and lower values at night due to decreased activity.The mid-day decreases are largest for NO x because of its short photochemical lifetime, and smallest for PM 10 , likely due to photochemical formation of secondary aerosols.The relatively large evening PM 10 peak occurs 1-2 h earlier than the CO and NO x secondary peaks, and may be due to temporal overlap between evening rush hour emissions and remaining secondary aerosols produced during the daytime, wind-related dust in the late afternoon (de Foy et al., 2008) or biomass burning plumes advected into the basin from surrounding areas late in the day (Moffet et al., 2007).Ozone concentrations increase rapidly during the late morning when photochemical radical production rates are largest (Volkamer et al., 2007) and peak in the early afternoon.
The weekend effect for CO, NO x , and PM 10 is seen clearly in Fig. 1 with the smaller morning peaks on Saturday and Sunday, compared to workdays (Monday-Friday).Early afternoon values are similar on workdays and Saturday, but distinctly lower on Sundays.Increases in CO and NO x are seen in the late evening on Friday and Saturday and persist into the early hours of the following day, as expected from increased weekend evening activities.For O 3 , a much smaller weekend effect, if any, is seen with values on Saturday and Sunday as high as those on workdays, and (as discussed below) occasionally even higher.The evenings of Friday and Saturday, and the early hours of the following day, have somewhat lower O 3 than on other nights, consistent with the higher NO x levels and O 3 loss by the reaction NO+O 3 →NO 2 +O 2 .Also notable is the earlier rise in O 3 concentrations on Sunday morning relative to other days, resulting from the earlier time that O 3 concentrations exceed those of NO, i.e. an earlier NO−O 3 cross-over as already seen in other studies, e.g., in Azusa, California (Fujita et al., 2003).The long-term behavior is shown in Fig. 2, where the morning maxima in CO, NO x , and PM 10 , and the afternoon maximum in O 3 are given for Wednesday and Sunday, averaged over all stations.Average CO values decreased sharply in the early 1990s following the closing of a major industrial facility in the city, and continued to decline most likely due to reductions in traffic-related emissions (Molina et al., 2002b).NO x and PM 10 values have decreased some since the beginning of the record but show little or no change in the last decade.Ozone values peaked in the early 1990s and continue to decrease.Lower values are seen on Sunday relative to Wednesday for CO, NO x , and PM 10 , but not for O 3 .Table 2 compares the workday averages with Saturday and Sunday values.For CO, NO x , and PM 10 , Saturday values generally fall between the workday and Sunday values, while for O 3 they are frequently highest (on 9 out of the 22 years).Workday O 3 was higher than either Saturday or Sunday for only 5 of these years, and not since 1994.
The detailed weekly patterns are shown in Fig. 3, averaged separately for each city sector (CT, NE, NW, SW, SE) over 1986-1992, 1993-2000, and 2001-2007.Considerable variation is noted by sector, even for the same years.The SW sector is particularly interesting, with relatively low morning CO, NO x , and PM 10 but high afternoon O 3 concentrations, indicating substantial contributions from advection during photochemical hours from other sectors, in agreement with the frequent "O 3 -South" episodes described by deFoy et al. (2005) and the confluence lines discussed by Cruz Nuñez and Jazcilevich Diamant (2007).Nevertheless, values of CO, NO x , and PM 10 are consistently lower on Saturday and more so on Sunday compared to the other days of the week, while no such reductions are seen in O 3 , except in the SW sector and there only during the earlier years.Variations between workdays are much less prominent, with some indication of increases of CO, NO x , and PM 10 in the early part of the week (Monday to Thursday) but with considerable variability, in agreement with meteorological studies that indicated nearly complete ventilation of the basin on a daily basis, with little day-to-day accumulation of pollutants (e.g., Fast and Zhong, 1998;de Foy et al., 2008 The amplitudes of the weekend effect, derived from the data shown in Fig. 3 using the harmonic regression described in Sect.2, are shown in Figs. 4 and 5.For CO, the amplitude (ppm) has decreased in approximate proportion to the decrease in average concentrations (see Fig. 2), so that on a relative basis (%) the weekend reductions have remained relatively constant at 40-50%.Relative reductions in NO x have also remained relatively constant, ranging between 40 and 60% in the last decade, while the PM 10 weekend effect amplitude is variable between 10 and 40%.In contrast, the O 3 weekend effect amplitude shows a positive trend, with values in the −20 to 0% range in the late 1980s, increasing to 0 to +10% in the last few years.This long-term positive trend for O 3 , coupled with the relative constancy of NO x , CO, and PM 10 relative weekend effect, has important implications for understanding the VOC-NO x -UV regime of México City's photochemistry, as will be discussed below.Some variations between the different urban sectors are seen in Figs. 4 and 5 but the qualitative features of the weekend effect are present in all sectors and are quantitatively more similar in recent years.
Seasonal variations are influenced by the dry (November-March) and wet (May-September) seasons.For 2001-2007 (Fig. 6) concentrations of CO, NO x , and PM 10 were largest during January and February, while O 3 peaked in March and April when solar actinic fluxes are higher.Lower values during the wet season are understood in terms of convective ventilation and wet removal.The relative amplitudes of the weekend effect show complex seasonal behavior (Fig. 7) for Fig. 3. Weekly patterns of the concentrations of CO, NO x , and PM 10 in the morning (average of the three highest concentrations between 7 a.m. and 12 noon) and O 3 in the afternoon (average of the three highest concentrations between 11 a.m. and 5 p.m.), by city sector (see legend).Averages are given for the time periods 1986-1992 (red), 1993-2000 (green), and 2001-2007 (blue).
reasons that are not entirely clear but may include a number of factors such as holiday activities (December) and regional influence from biomass burning in the late dry season (March-May).For 1986For -1992For and 1993For -2000 (not shown) (not shown), the general seasonal patterns were similar with concentrations of CO and NO x peaking in January and February, O 3 peaking in May, and no clear seasonal trend of the relative weekend changes.

Hypotheses for the weekend effect
México City's surface observations show a definite pattern over weekly periods: CO, NO x , and PM 10 morning concentrations are smaller on weekends relative to workdays, by ca.40-50%, 40-60%, and 10-40%, respectively; O 3 afternoon weekend concentrations are not much smaller, and are sometimes even larger, than the workday values, with differences increasing from −20 to 0% in the late 1980s, to 0 to +10% in the past decade.These observations of the weekend effect offer the opportunity to better understand the chemical regime responsible for the formation of O 3 .The central issue is to explain why O 3 concentrations remain relatively unchanged on weekends, relative to workdays, when precursor emissions are considerably lower.Lawson (2003) summarized the possible reasons in terms of six hypotheses: (1) Lower weekend NO x emissions, leading to less NO x inhibition of O 3 formation if under VOC-limited conditions, (2) later timing of NO x emissions on weekends, (3) carryover of previous day pollutants at the surface, (4) carryover of previous day pollutants aloft, (5) higher weekend VOC emissions, and (6) higher weekend photolysis frequencies due to less aerosol.The first hypothesis, that workday O 3 production is VOClimited and NO x -inhibited, appears to be the most plausible explanation for the observed weekend effect in México City.The sensitivity of O 3 production to VOC changes is always positive (albeit small at low NO x ), while it can be either positive or negative with respect to NO x changes, the negative values representing NO x inhibition of O 3 production in the VOC-limited regime.In this regime, hypothetical reductions in only VOC emissions would lead to lower O 3 , while equally hypothetical reductions in only NO x emissions would lead to higher O 3 .The near equality of workday and weekend O 3 then arises from the simultaneous decreases in VOC and NO x emissions and their opposing effects on the O 3 production rates.
It is important to note that direct VOC measurements were not used in our analysis.Such measurements for México City are relatively sparse and from only a few locations (e.g.Blake and Rowland, 1995;Raga et al., 2001;Arriaga-Colina et al., 2004;Velasco et al., 2007;and references therein).The spatial and temporal variability of the weekend effect is rather large even within the much more comprehensive CO data set (e.g.Fig. 3), and would be much more difficult to quantify with the limited available VOC record.On the other hand, VOCs are several times more reactive (with respect to OH radicals) than CO in México City (see, for example, Fig.  of Madronich, 2006), so an open issue is whether variations in CO can be used as a proxy for variations in VOC reactivity.Mexico City's mobile sources account for ∼98% of CO emissions, but only ∼40% of VOC emissions with the balance mostly from area sources such as solvent use and painting (Molina et al., 2002a).Whether the emissions from these area sources decrease on weekends by a similar fraction as mobile sources is uncertain.Some support for this comes from observations of robust CO vs. VOC correlations during the MILAGRO field campaign (deGouw, J. et al., in prep., 2008), as well as measurements in Southern California showing similar relative workday to Sunday reductions by 16-30% for VOCs and 12-32% for CO (Blanchard and Tanenbaum, 2003).
The other hypotheses (2-6) for explaining the weekend effect are not supported by the observations.Timing of the NO x emissions (hypothesis 2) is not very different on weekend mornings than on workdays (see Fig. 1).Similarly, Marr and Harley (2002) showed that change in timing of emissions is only a minor contributor to the weekend effect in Central California.Carryover of pollutants from the previous day (hypotheses 3 and 4) is small, as can be seen in Fig. 3, consistent with meteorological studies suggesting nearly complete daily ventilation of the basin (e.g.deFoy et al., 2008).The possibility of higher weekend VOC emissions (hypothesis 5) has been examined for California where outdoor cooking and lawn mowing are common weekend activities, but even there it was not supported by detailed emissions inventories (Chinkin et al., 2003); it seems equally unlikely for México City given the large weekend decrease in CO.The workday to weekend increase in photolysis frequencies www.atmos-chem-phys.net/8/5313/2008/Atmos.Chem.Phys., 8, 5313-5325, 2008 (hypothesis 6), owing to the heavier workday aerosol loading, merits some consideration.Castro et al. (2001) showed that surface NO 2 photolysis frequencies (J NO2 ) were reduced in México City by 20-30% compared to outside the city, and more recent measurements during the MILAGRO campaign show comparable reductions in actinic fluxes at ultraviolet wavelengths (Madronich et al., in prep., 2008).Weekend reductions in PM 10 are seen to be in the range 10-40% (Fig. 3), which if applied to the J NO2 reductions found by Castro et al.
give an outside range of weekend enhancement of photolysis rates between 2% and 12% at the surface.Vertically averaged values in the PBL would be expected to be somewhat smaller, so that the resulting enhancement in O 3 production is small although not negligible.

Evaluation of L N /Q
We consider here whether the magnitudes of the observed weekend changes in CO, NO x , and O 3 are consistent with photochemical understanding.In Sect.2.2 the theoretical expectation of how these magnitudes are related to each other was derived (Eq.7) within the approximations discussed there.The algebraic form of Eq. ( 7) permits any negative or positive value of L N /Q (from −∞ to +∞) for independently selected combinations of δCO, δNO x , δJ , and δO x .However, the photochemical interpretation of L N /Q, as the fraction of radical termination effected by NO x chemistry, limits its possible values to the range 0-1.The question then is whether the observed weekend effect values of δCO, δNO x , δJ , and δO 3 are consistent with this chemical interpretation.Figure 8 shows the L N /Q values calculated from the observed CO, NO x , and O 3 changes (taken from Table 2), approximating δO x by δO 3 and with a photolysis enhancement (δJ =0.07) in the mid range of values discussed above (sensitivity to these approximations is discussed in Sect.4.3).The values are rather scattered but clearly fall near or within the chemically permissible range, and moreover are generally between 0.5 and 1.0 as expected for a VOC-limited regime; the workday-Saturday values are somewhat higher that Saturday-Sunday values as expected from more intense NO x inhibition on workdays; and a slight upward trend in L N /Q is seen, especially for the last decade, as expected from the decreasing trend in concentrations of CO (and presumably VOCs).However, such small variations should be viewed with caution, because the uncertainty in L N /Q is about ±30%, as estimated by error propagation in quadrature through Eq. ( 7) of the standard deviations in δCO, δNO x , and δO 3 (ca.10%, 10%, and 7%, respectively, from Fig. 5).The L N /Q values discussed so far were based on the average of all days for which data were available, and it is not obvious a priori that VOC limitation and NO x inhibition persist also for very high O 3 episodes.To test this, we selected the upper 75th percentile having the highest ozone concentrations (i.e.25% of days, separately for workdays, Saturdays, and Sundays) and recalculated L N /Q for this subset.Figure 8 shows that values of L N /Q for the high O 3 days are still in the VOC-limited and NO x -inhibited range for most years, although with more scatter and an anomalous value for 1991 probably due to large intra-annual emission changes as already mentioned in Sect.3. It should be noted that selecting a subset of the days introduces additional scatter and possible bias because: (1) Sample size is reduced, e.g. for the upper 75% percentile only 13 weekends are available per year.(2) While yearly averages include all days, a subset may sample workdays and weekends from different weeks, thus amplifying variability from seasonal dependences.(3) When the selection is made on the basis of high O 3 , some days with high NO x may be excluded precisely because O 3 formation is NO x inhibited.This bias is more frequent on workdays because they are more strongly NO x -inhibited (indeed, for 2007 Saturday NO x values were actually higher than for workdays for the 75th percentile O 3 subset, while for all-day averages, shown in Table 2, Saturday NO x is lower as expected).
The measurement-based values of L N /Q found here are supported by recent modeling of the April 2003 MCMA field campaign.Lei et al. (2007Lei et al. ( , 2008) ) analyzed the correlations between simulated radical sources and production rates of NO z (= NO y −NO x ).The slopes of these correlations are equivalent to L N /Q and show that, during the afternoon in the Mexico City urban area, over 90% of the radicals are removed via NO x chemistry.
One possible confounding factor is that NO x and VOC emissions in one part of the city may be transported over a few hours by urban scale circulations to produce high O 3 concentrations in other parts of the city, under some specific meteorological conditions as noted by de Foy et al. (2005).Our use of city-wide averages evidently smoothes over such spatial variations, and in any case the weekend effect was noted to be qualitatively similar in all city sectors (see Figs. 4  and 5), so it is unlikely that such circulations would alter our conclusion about VOC-limitation.
Another interesting result is the detection of a long-term positive trend in the O 3 weekend effect, while the CO and NO x weekend fractional reductions have remained essentially constant (see Fig. 5 for concentrations, or Fig. 8 for L N /Q).This is associated with the long-term decrease in CO concentrations, presumably correlated with decreases in VOC concentrations, while NO x concentrations have remained largely unchanged.A decrease in the VOC/NO x ratio implies a shift toward more VOC-limited conditions over the decades examined here.Earlier studies using threedimensional chemistry-transport models (CTMs) suggested a NO x -limited regime (Molina et al., 2002a;West et al., 2004), while more recent CTM studies indicated a VOC-limited regime (Tie et al., 2007;Lei et al., 2007Lei et al., , 2008)).It has been so far unclear whether this discrepancy is due to improvements in the models, or to changes in the actual emissions.Our observation of a long term positive trend in the O 3 weekend effect provides at least a partial explanation for the different modeling results, suggesting a more VOC-limited regime for the recent years.It should be cautioned, however, that our use of CO as a proxy for VOC reactivity may be less valid over very long time periods, because of possible long-term changes in the detailed speciation of the many components that make up the reactive VOC mixture.We note also that  urban and regional development has increased greatly in the past two decades (Lezama et al., 2002), so that the monitoring stations may have been sampling a more urban chemical regime in recent years.

Sensitivity of L N /Q
The sensitivity of estimated L N /Q to several assumptions is presented here.For reference, we used the workday to Sunday changes, average of all days, and weekend photolysis enhancement of 7% (as in Fig. 8).We first consider the possibility that the weekend enhancements in photolysis frequencies may have been as large as 12% (δJ =0.12 rather than 0.07).Figure 9 shows that the estimated values of L N /Q are then smaller by about 0.05.The net production of O 3 is usually photon-limited (in all but the most pristine parts of the troposphere) so that weekend enhancements in J -values contribute to the persistence of high O 3 values, and less change in NO x inhibition is needed to explain the observations, leading to smaller values of L N /Q.Although the values of L N /Q do remain mostly in the VOC-limited regime (>0.5) even with these larger J -value enhancements, the sensitivity is seen to be significant and emphasizes the need for accurate long-term observations of the urban ultraviolet environment.The sensitivity to using O x rather than O 3 in the analysis is shown in Fig. 9 was then used in Eq. ( 7) in place of δO 3 .For the workday to Sunday differences, this reduces the values of L N /Q by 0.04-0.13over the data record, as could be expected from Fig. 1 which shows that values of NO x in the early afternoon (mostly NO 2 ) are lower on Sunday than on other days.Although O 3 on Sundays is typically the same or even slightly higher than on workdays (Table 2), the total O x is slightly lower due to the lower NO 2 .Even with this correction, L N /Q is still within the VOC-limited regime.For Saturdays (not shown) corrections to L N /Q are negligible because NO x values on Saturday afternoons are nearly identical to those on workdays (see again Fig. 1), so δO x is well approximated by δO 3 .Finally, Fig. 9 shows the sensitivity of L N /Q to an assumed 40 ppb of background O 3 .The derivation of L N /Q (Eq.7) refers exclusively to O x produced during the same day and does not account for any O 3 that may have been present in the atmosphere from production in previous days.The amount of background O 3 on any particular day is not well known, but ozone sondes (Thompson et al., 2008) showed concentrations above the PBL in the range of 30-50 ppb during March 2006 and 40-60 ppb during August-September 2006.These sondes were launched in the early afternoons and may reflect some same-day production in addition to background O 3 .Thus our use of 40 ppb is probably a reasonable estimate for this sensitivity study.Figure 9 shows that the effect of background O 3 is negligible when L N /Q values are higher than ∼0.8, as in the recent years, but could lead to overestimation of L N /Q by as much as 0.1 when the values are lower.

Conclusions
México City experiences a weekend effect in its air quality similar to that found in many cities around the world: Although concentrations of O 3 precursors NO x , CO, and (presumably) VOCs are significantly lower on Saturday and even more so on Sunday compared to workdays, the concentrations of O 3 change only minimally, and in some cases are even larger.This effect has become more pronounced in recent years because of significant emission reductions of CO and VOCs but relatively steady NO x emissions.The observed weekend effect is consistent with a VOC-limited, NO x -inhibited chemical regime for O 3 production during workdays.Nitrogen chemistry accounts for most of the radical loss, with L N /Q values (from Figs. 8 and 9) in the range 0.75 to 0.95; with these values, the normalized sensitivities of O 3 production (Eqs.1-3) lie in the range −0.25 to −0.8 for NO x , 0.6 to 0.9 for VOCs, and 0.8 to 0.95 for photolysis frequencies.In this regime, any magnitude of reduction in VOC emissions would contribute to lowering ambient O 3 concentrations, while only large reductions in NO x emissions would prove effective, with smaller incremental reductions being ineffective and possibly even detrimental by increasing local O 3 production, depending on specific location and time.
There are of course many other reasons for reducing NO x emissions.NO 2 is per se an important pollutant, and many nitrogen-containing compounds formed in the atmosphere are noxious, e.g.nitric acid, peroxy acyl nitrates (PANs), and nitro-cresols.Furthermore, the NO x inhibition of O 3 production is likely temporary, and by slowing the oxidative reactivity it allows more yet-to-be-reacted O 3 precursors to be exported from the city to the regional scale, including slower-reacting hydrocarbons and partly oxygenated VOCs.Many organic nitrogen species (e.g., alkyl nitrates and PANs) formed in the urban atmosphere have relatively long lifetimes and can, through later thermal or photolytic decomposition, be an important source of NO x to the regional and global atmosphere where O 3 production is generally NO x -limited.
This analysis was confined to the urban network of monitoring stations for which long term measurements are available, and is therefore only valid for the geographic area which these stations represent.Over the past two decades, urban expansion beyond the monitored area and suburban development make it important to understand at which point the chemical regime transitions from VOC-limited to NO xlimited.While this can be achieved by expansion of the long-term monitoring network, it can also be addressed by improved numerical models that have been evaluated with observations in both urban and regional chemical regimes.

Fig. 4 .
Fig. 4. Amplitude (absolute concentration) of the weekend effect for CO, NO x , PM 10 , and O 3 .Thick line is the average of all stations, while individual thin lines (legend in lower right panel) give results by sector. 3

Fig. 5 .
Fig. 5. Amplitude (relative, %) of the weekend effect for CO, NO x , PM 10 , and O 3 .Thick line is the average of all stations, while individual thin lines (legend in lower right panel) give results by sector.

Fig. 6 .
Fig. 6.Seasonal variation of the concentrations of CO, NO x , and PM 10 in the morning (average of the three highest concentrations between 7 a.m. and 12 noon) and O 3 in the afternoon (average of the three highest concentrations between 11 a.m. and 5 p.m.), for the years 2001-2007.

Table 1 .
Number of station-days reported in each city sector

Table 2 .
Concentrations of CO, O 3 , NO x , and PM 10 in México City on workdays (M-F), Saturday (Sat), and Sunday (Sun); 3-h daily maxima a averaged over all stations and days of year.
a Average of each day's three highest values between 7 a.m. and noon for CO, NO x , and PM 10 , and between 11 a.m. and 5 p.m. for O 3 .
Fraction of radical loss by NO x chemistry relative to total radical loss (L N /Q) derived from the observed weekend changes in CO, NO x , and O 3 concentrations.Black line with solid circles is for workday (Monday-Friday) to Sunday changes; red line with open squares is for workday to Saturday, blue line with open triangles is for Saturday to Sunday changes; dashed black line with open circles is for workday to Sunday changes but including only high O 3 days (75th percentile).All calculations were made with a 7% change in photolysis rates (δJ , see text).
. Co-located simultaneous measurements of NO 2 and O 3 were summed to compute δO x which