The production and persistence of ΣRONO 2 in the Mexico City plume

Abstract. Alkyl and multifunctional nitrates (RONO2, ΣANs) have been observed to be a significant fraction of NOy in a number of different chemical regimes. Their formation is an important free radical chain termination step ending production of ozone and possibly affecting formation of secondary organic aerosol. ΣANs also represent a potentially large, unmeasured contribution to OH reactivity and are a major pathway for the removal of nitrogen oxides from the atmosphere. Numerous studies have investigated the role of nitrate formation from biogenic compounds and in the remote atmosphere. Less attention has been paid to the role ΣANs may play in the complex mixtures of hydrocarbons typical of urban settings. Measurements of total alkyl and multifunctional nitrates, NO2, total peroxy nitrates (ΣPNs), HNO3 and a representative suite of hydrocarbons were obtained from the NASA DC-8 aircraft during spring of 2006 in and around Mexico City and the Gulf of Mexico. ΣANs were observed to be 10–20% of NOy in the Mexico City plume and to increase in importance with increased photochemical age. We describe three conclusions: (1) Correlations of ΣANs with odd-oxygen (Ox) indicate a stronger role for ΣANs in the photochemistry of Mexico City than is expected based on currently accepted photochemical mechanisms, (2) ΣAN formation suppresses peak ozone production rates by as much as 40% in the near-field of Mexico City and (3) ΣANs play a significant role in the export of NOy from Mexico City to the Gulf Region.


Introduction
The chemistry of alkyl and multifunctional nitrates acts to suppress O 3 formation in the near field of urban plumes and then to extend the range of ozone formation in the far field by releasing NO x in locations far from NO x emissions (e.g. Ito et al., 2007). However, there are few detailed observational tests capable of assessing the quantitative importance of these effects and current models incorporate contradictory assumptions about elements of the chemistry that are not well constrained by lab or by field observations. Recent analyses of models and their differences (Wu et al., 2007;Ito et al., 2007), field observations (Giacopelli et al., 2005;Farmer and Cohen, 2007;Horowitz et al., 2007;Perring et al., 2009a) and laboratory measurements (e.g. Paulot et al., 2009) have focused attention on the extent to which RONO 2 molecules preserve the −ONO 2 functional group upon oxidation. An emerging theme from these papers is that the lifetime of total RONO 2 (denoted ANs hereafter), rather Published by Copernicus Publications on behalf of the European Geosciences Union.

7216
A. E. Perring et al.: The production and persistence of RONO 2 in the Mexico City plume than the lifetime of any individual RONO 2 molecule, is the key to understanding the effects of RONO 2 formation on atmospheric chemistry.
ANs have been observed to be a significant fraction of NO y in a number of different chemical regimes (Day et al., 2003;Rosen et al., 2004;Cleary et al., 2005;Perring et al., 2009a) and have been inferred to be an important photochemical product in Mexico City with potential concentrations of several ppb (Dunlea et al., 2007). Here we present observations of ANs, measured by thermal dissociation coupled to laser-induced fluorescence detection of NO 2 (Thornton et al., 2000;Day et al., 2002), in and downwind of Mexico City using the NASA DC-8 platform during the Megacity Initiative: Local and Global Research Observations (MILAGRO) phase of the INTEX-B campaign in the spring of 2006 (Singh et al., 2009;Molina et al., 2010). The aircraft flights targeted plume evolution and crossed the Mexico City plume at its origin and at distances as far as 1000 km downwind. We use these observations to explore the role of ANs as they affect ozone and nitrogen oxides. We show that AN chemistry has important consequences for urban O 3 , for regional photochemistry and for the evolution of O 3 in the Mexico City plume. We examine aspects of the chemistry that are specific to this plume. We investigate the progression of the NO y distribution as the plume ages with special attention to the continued significance of ANs as a fraction of NO y .

Measurements
Observations described here were made aboard the NASA DC-8 during the Intercontinental Transport Experiment-Phase B (INTEX-B), which took place in the spring of 2006. INTEX-B, a NASA-led multi-partner atmospheric chemistry campaign, has been described elsewhere (Singh et al., 2009). One of the stated goals was to investigate the extent and persistence of Mexican pollution outflow as part of the MILAGRO campaign (Molina et al., 2010).
NO 2 , PNs, and ANs were measured using the Berkeley thermal dissociation-laser induced fluorescence instrument (Thornton et al., 2000;Day et al., 2002). Briefly, gas is pulled simultaneously through four channels consisting of heated quartz tubes maintained at specific temperatures for the dissociation of each compound class above. Each heated section is followed by a length of PFA tubing leading to a detection cell where NO 2 is measured using laser-induced fluorescence. Due to differing X-NO 2 bond strengths, PNs, ANs and HNO 3 all thermally dissociate to NO 2 and a companion radical at a characteristic temperature. The ambient channel measures NO 2 alone, the second channel (180 • C) measures NO 2 produced from the dissociation of PNs in addition to ambient NO 2 so the observed signal is NO 2 + PNs, the third channel (380 • C) measures NO 2 + PNs+ ANs, and the last channel (580 • C) measures NO 2 + PNs+ ANs+HNO 3 . Concentrations of each class of compound correspond to the difference in NO 2 signal between two channels set at adjacent temperatures. The difference in NO 2 signal between the 180 • C and the 380 • C channel, for example, is the ANs mixing ratio. The instrument deployed for INTEX-B had a heated inlet tip that split in two immediately. Half of the flow was immediately introduced to heated quartz tubes for detection of ANs and HNO 3 while the other half was introduced to an additional heated quartz tube for detection of PNs and an ambient temperature channel for detection of NO 2 .
Ambient NO 2 and NO 2 produced by thermal dissociation was observed by laser-induced fluorescence as described in detail by Thornton et al. (2000). Briefly, a tunable dye laser is pumped at 7 kHz by a Q-switched, frequency doubled Nd +3 YAG laser. The incoming gas is cooled through the use of a supersonic expansion (Cleary et al., 2002) and the dye laser, utilizing Pyrromethene 597 in isopropanol, is tuned to an isolated rovibronic feature of jet-cooled NO 2 at 585 nm. The frequency is held for 20 s at the peak of this feature and then for 5 s at an offline position in the continuum absorption. The ratio of peak to off-resonance fluorescence of the chosen feature is 10:1 at 1 atm and the difference between the two signals is directly proportional to the NO 2 mixing ratio. The laser light is focused in series through two multi-pass (White) cells (discussed in more detail below) and the redshifted fluorescence is detected using a red-sensitive photomultiplier tube (Hamamatsu). Fluorescence counts are collected at 5 Hz, scattered light at wavelength less than 700 nm is rejected by band-pass filters and time-gated detection is used to eliminate noise resulting from scattered laser light in the cell. We observe a dependence of NO 2 fluorescence on the external pressure. We calibrate the NO 2 LIF vs. altitude by direct measurement of NO 2 from a standard addition during a test-flight. Calibrations were performed at least once every two hours during a level flight leg using a 4.7 ppm NO 2 reference gas (Praxair) with a stated certainty of ±5%. The reference gas was compared to a library of standards in lab both before and after the campaign. The individual standards are compared on a regular basis (about every 6 months) to ensure stability and highlight when a given tank has degraded. These standards have been observed to remain stable for up to 5 years and to be accurate at atmospherically relevant mixing ratios to within 1% (Bertram et al., 2005).
The instrument deployed for INTEX-B had two detection cells. The direction of flow into the cell was controlled using a three-way valve and a bypass pump was used to maintain flow in the non-sampled channel. Cell 1 sampled either the ambient (75% of the time) or the 380 • C channel (25% of the time) while cell 2 sampled either the 180 • C (50% of the time) or the 580 • C channel (50% of the time). Thus for every 2 min duty cycle there were three 20 s average measurements of NO 2 , two 20 s average measurements of PNs, one 20 s average measurement of ANs, one 20 s average measurement of HNO 3 and one 20 s average measurement of the sum ( PNs+ ANs+HNO 3 ). As the ANs measurement is a subtraction, the uncertainty depends both on ANs and on the sum (NO 2 + PNs). For example, if there were 100 ppt each of NO 2 and PNs, the precision of the ANs measurement would be ∼15 ppt in 20 s for S/N=2 at typical laser power and cell alignment. If there were 1 ppb each of NO 2 and PNs the precision of the ANs would be 40 ppt.
We expect to detect both gas and aerosol phase organic nitrates with the TD-LIF instrument because the analogous TD-LIF measurement of HNO 3 has been shown to be the sum of aerosol and gas-phase HNO 3 (Fountoukis et al., 2009). While a direct intercomparison of the ANs measurement has not been published, we have sampled pure standards of ethyl nitrate, propyl nitrate and isoprene nitrates (synthesized by wet chemical methods in the laboratory) in air. In each case we observe signal only in the ANs channel of the TD-LIF indicating that the nitrates are not dissociating in the other temperature channels. Comparison of TD-LIF observations of an isoprene nitrate standard to observations made using a PTR-MS show both instruments to be consistent to within 10% (Perring et al., 2009b). Comparisons of NO 2 and PNs have also been described and indicate similar or better accuracy for these species Fuchs et al., 2010;Wooldridge et al., 2010).
For the present analysis, we use the HNO 3 measurement made by the University of New Hampshire with a mist chamber followed by ion chromatography Dibb et al., 2006). This measurement represents the sum of gas-phase HNO 3 and fine (submicron) aerosol nitrate. Hydrocarbons and C 1 -C 5 alkyl nitrates were measured by UC Irvine using gas chromatography of whole air samples (Colman et al., 2001). Oxygenated volatile organic carbon species (methyl-ethyl-ketone, methanol, ethanol, acetone and acetaldehyde collectively referred to, when combined with CH 2 O, as oxidized volatile organic carbon or OVOC) were measured by NASA Ames using gas chromatography (Singh et al., 1999). NO (Georgia Tech) and O 3 (NASA Langley) were measured through chemiluminesence. OH and HO 2 were measured by laser-induced fluorescence by Penn State (Faloona et al., 2004). CH 2 O was measured by NCAR using tunable diode laser absorption spectroscopy (TDLAS) (Fried et al., 2003) and by the University of Rhode Island (URI) using an enzyme-derivatization fluorescence technique following collection in an aqueous medium and high performance liquid chromatographic analysis (Heikes, 1992). There is an unresolved ∼18% discrepancy between the two measurements so we have used an average of the two in the present analysis. URI also employed the enzyme-derivatization fluoresence technique to measure hydrogen peroxide (H 2 O 2 ). CO was measured by differential absorption as described by Sachse et al. (1987). NO y was not measured explicitly but for the purposes of this manuscript we consider it to be the sum of all measured reactive nitrogen species (NO, NO 2 , PNs, ANs and HNO 3 ). Photolysis frequencies are calculated from spectraradiometer measure-ments as described by Shetter and Muller (1999). Ten-day back trajectories from locations of the DC-8 were calculated using the National Weather Service's Global Forecast Model (GFS) analyses of basic parameters as described by Fuelberg et al. (2007). The GFS data were available at 6 h intervals on a 1 • latitude/longitude grid at 64 vertical levels.
The present work uses data from a 1-min merge available at http://www-air.larc.nasa.gov/missions/intex-b/intexb. html. We use back trajectories provided with the merged chemical data to select measured air masses that passed within ∼160 km (1.5 • ) of the T0 site in the center of Mexico City at pressures higher than 680 mbar. The elevation of Mexico City is 2240 m and typical surface pressure is ∼770 mbar. 680 mbar corresponds to an elevation of about 1 km a.g.l. (above ground level). Considering the 6 local flights out of Houston, there are a total of 2591 trajectories and, of those, 422 satisfy our criteria of having passed through the boundary layer in the vicinity of the Mexico City Metropolitan Area (MCMA).

Results and analysis
The formation of alkyl and multifunctional nitrates ( RONO 2i ≡ ANs) occurs as a result of reactions between organic peroxy radicals (RO 2 ) and NO (Reaction R1a). The alternative to alkyl nitrate formation is propagation of the radical chain of events, conversion of NO to NO 2 , and subsequent ozone production (Reaction R1b).
The ratio k 1a /(k 1a +k 1b ) is α, the nitrate branching ratio. In this analysis, we examine the concentration of ANs and the fraction of NO y that is represented by ANs as a function of time since emission of NO x and VOC at the point of origin in downtown Mexico City. We discuss the evolving relationship between ANs and O x Sect. 3.1 we develop a measure of airmass age. In Sect. 3.2 we describe the temporal evolution of NO y partitioning in the plume, in Sect. 3.3 we describe the temporal evolution of the correlation of AN with O 3 in the plume.

Photochemical age
We use the ratio of 2-butyl nitrate (CH 3 CHONO 2 CH 2 CH 3 ) to butane, an indicator of time since last anthropogenic influence, to sort the MCMA plume points by increasing photochemical age. This indicator of age has been discussed previously by Bertman et al. (1995). Photochemical age indicators depend on the assumption that parent and daughter molecules, in this case a particular straight-chain alkane and its daughter alkyl nitrate arise exclusively from a single chemical reaction (or a chain with a well-defined rate limiting step) and that the loss processes of the daughter are slower than the parent and well known. In addition, this analysis assumes that the emissions are effectively from a single isolated point source (LaFranchi et al., 2009). Depending on the accuracy required, mixing of a background into the plume must also be taken into account (Day et al., 2003). Neglecting mixing into a constant background doesn't affect the time ordering of the age indicator but does make its absolute magnitude less accurate. In this case we choose butane and 2-butyl nitrate as the parent-daughter pair. Ages calculated using ratios of 2-and 3-pentyl nitrate to pentane give similar results. Neglecting mixing, and assuming that every RO 2 formed reacts with NO, the change in the concentration of a nitrate over time is then: If we assume [RONO 2 ] 0 =0 and solve for t we find: Bertman et al. (1995) showed that these assumptions were adequate for evolving airmasses observed at Scotia Pennsylvania and the Kinterbish Wildlife Area, Alabama when applied to nitrates derived from hydrocarbons larger than propane. Figure 1 shows the distance from T0 (panel a), the observed ratio of 2-buylnitrate to butane (panel b), the age calculated from the 2-butyl nitrate to butane ratio (panel c) and the observed NO x to HNO 3 ratio (panel d) from a segment of the DC8's flight of 16 March 2006. The NO x to HNO 3 ratio, which has also been widely employed as an indicator of photochemical processing, is expected to decrease with increased age as long as oxidation of NO 2 is more rapid than Locations of data that trace back to the region around Mexico City colored by their photochemical age in hours of processing at OH=3×10 6 molecules/cm 3 . The black star represents the location of Mexico City and the dashed black circle shows the limit of the region we have defined as the Mexico City Area (within 100 km of the city center). The colored circles represent distances from the center of Mexico City of 200 km (blue), 500 km (green) and 1000 km (red). deposition of HNO 3 . Figure 2 shows the calculated age of all points that have back trajectories that trace back to Mexico City and Fig. 3 shows the relationship between the calculated age for these points and the distance to the T0 site in downtown Mexico City. Although we use a single value for OH in these calculations, it should also be noted that the rate of photochemical aging slows as the plume becomes more dilute and the OH concentration decreases. The OH concentration drops by more than a factor of 2 between the center of Mexico City and the Gulf of Mexico and this drives the shape of the curve shown in Fig. 3.
Much of the spread in the relationship between photochemical age and distance to the T0 site can be explained by variation in wind velocity. For example, the average measured wind velocity below 4 km altitude (where most of the plume sampling occurred) on 16 March was ∼5.5 m/s (blue points) while that on 19 March was ∼10 m/s (red points). The higher wind velocity on 19 March carried the airmass farther from the source at equivalent photochemical aging than the mean. Note that the choice of OH concentration of 3×10 6 molecules/cm 2 gives photochemical ages in the first two days of plume evolution that are similar to the transport times calculated from observed wind speeds; i.e. on the day with a windspeed of 5.5 m/s the calculated photochemical age of the points near 1000 km is ∼45 h while the transport time would be ∼50 h. Additional spread in the transport time-photochemical time relationship is also introduced as a result of using a photochemical clock that turns off at night. Figure 4a shows concentrations of the components of NO y as a function of photochemical age and 4b shows NO y speciation as a function of photochemical age. Near Mexico City (time < 6 h) NO x is ∼3.5 ppb and the dominant component of NO y . Within 200 km (by ∼20 h) NO x decreases to <25% of NO y and at the longest time considered (130 h) NO x is 15% of NO y , comparable to PNs and ANs. HNO 3 is a minor fraction of NO y (10%) close to Mexico City and increases to ∼60% at the longest times.

NO y speciation in the Mexico City Plume
The concentration of PNs peaks at ∼8 h and PNs as a fraction of NO y peak at ∼15 h (100 km) where they account for 30% of NO y after which they decrease to about ∼15% of NO y at ∼40 h. Qualitatively this is the expected pattern. Previous analyses have shown that net production of peroxy nitrates occurs in the near-field of source regions where concentrations of NO 2 and RO 2 are high and the formation rate exceeds the rate of thermal decomposition. As the plume becomes more dilute, the formation rate will decrease exponentially while the thermal decomposition rate is determined only by the temperature. For urban plumes that experience rapid lofting and wet removal of both HNO 3 and ANs, PNs become the dominant NO y reservoir in the colder, lofted airmass (Altshuller, 1993;Singh and Salas, 1989;Talbot et al., 2003). The Mexico City Plume, as selected based on the criteria described above, was encountered exclusively below 5 km. The plume was almost entirely Table 1. Column 1 identifies the hydrocarbons and columns 2 and 3 show OH rate constants and nitrate branching ratios for the suite of hydrocarbons observed in Mexico City. Columns 4-6 show the median concentrations and the associated ozone and alkyl nitrate production rates respectively.    (95%) at temperatures above 280 K corresponding to a PAN thermal lifetime of less than 10 h. ANs start at 10% of NO y near Mexico City. At the longest time they are of comparable importance to PNs (∼15% of NO y ). If ANs react and release NO 2 , they may be important in the redistribution of reactive nitrogen within the Gulf of Mexico.
We can account, at least partially, for dilution by comparing enhancements over the background to CO, which is a conserved tracer on the timescale of this plume study. CO is lost to reaction with OH at a rate of approximately 6% per day. Photochemical production of CO is also minimal on these timescales. We calculate that, over 40 h of photochemical aging at our assumed OH concentration, 2.6 ppb of CO would be produced from methane oxidation and an additional 3-4 ppb would be produced from oxidation of formaldehyde, acetone and ethane (using concentrations of each observed in Mexico City). Oxidation of larger hydrocarbons is likely to contribute primarily through the production of formaldehyde which we account for above. For comparison, CO decreases from 600 ppb at 0-5 h photochemical age to 450 ppb at about 10 h and is mixing into a background of 132 ppb CO. The expected CO production is less than 2% of the observed decrease in concentration over 40 h of photochemical aging and the decay in CO is therefore primarily a measure of dilution. Molecules that decay faster than CO can be considered to be removed by chemistry or deposition while those that decay more slowly are being produced in the plume. Figure 5a shows fractional enhancement over the background for NO y , CO, NO x , PNs and ANs. The fractional enhancement over background for species X is defined as: where X initial is the mean observed concentration of species X at less than 10 h of photochemical aging and X background is the mean observed concentration of species X at the longest photochemical ages over the Gulf of Mexico. Background concentrations used were 132 ppb for CO, 1.22 ppb for NO y and as shown in Fig. 4a for the different NO y species. The near identical fractional dilutions of NO y and CO indicates that the airmasses were not subject to large depositional losses, consistent with the fact that the plume was primarily encountered between 2 and 4 km and should have been connected to the planetary boundary layer weakly if at all. Figure 5b shows the fractional enhancement of the components of NO y divided by the fractional enhancement of CO, which should effectively cancel out the effect of dilution and allow us to examine the effect of chemistry. PNs, which are primarily produced within the first 15 hrs of plume evolution, are depleted relative to CO throughout the plume. NO x is initially enhanced relative to CO because net dissociation of PNs is a source of NO x . In the most aged airmasses considered, NO x is depressed relative to CO because it has been converted to HNO 3 . HNO 3 is chemically produced in the plume and thus lies above the dilution line. ANs behave similarly to HNO 3 confirming that they, too, continue to be produced as the plume ages although most of the production occurs early in the plume. The lines all converge at the longest photochemical age because that is what we have defined as "background" conditions.

The evolution of the relationship between ANs and O x
Since O x (O x =O 3 +NO 2 ) and ANs arise from alternative channels of Reaction (R1) the slope of the correlation of O x and ANs is a measure of the balance between chain propagation (O x production) and termination ( ANs production) as long as losses of both are slow relative to production. This assumption will be investigated in more detail in Sect. 4.2.
Lower slopes imply a more significant role for ANs formation. Figure 6 shows . It is therefore reasonable to assume a constant background for mixing with AN levels that are low (0-50 ppt) and O x that is 50-55 ppb. If the whole plume is being diluted into the same background mixture, then dilution will impact the observed concentrations of O x and ANs but not affect the slope of the correlation. The change in slope with time indicates a variable role for AN production over the lifetime of the plume. Slopes of O x / ANs have been reported for a number of different locations using ground-based AN measurements at UC-Blodgett Forest Research Center (Day et al., 2003), Granite Bay CA (Cleary et al., 2005), Houston, TX (Rosen et al., 2004), the Big Hill field site in the Sierra Foothills (Murphy et al., 2006) and using airborne measurements made over the southeastern US aboard the NASA DC-8 (Perring et al., 2009a). Typically lower ANs v. O x slopes are observed in urban locations than in more rural ones. For comparison with the current dataset, Rosen et al. (2004) found a slope of 29 O x / ANs in the morning in Houston which increased to 41 in the afternoon and the average afternoon slope that we observed at the T1 site in Mexico City during MILAGRO was 22. Levels of ANs in remote areas of the troposphere have been observed to be very low and observed O x / ANs slopes in the remote Pacific are found to be 200-500. This is higher than the slope found for even the longest times considered in the present analysis and illustrates the fact that the Mexico City plume impacts much of the Gulf region.
The instantaneous production rates of O x and ANs are given by: where α i is the nitrate branching ratio for RH i and γ i is the immediate number of O 3 produced from the oxidation of RH i . γ is equal to 2 for most hydrocarbons since Reaction (R1b) typically results in the net production of 2 O 3 molecules; one from photolysis of the NO 2 formed directly from RO 2 +NO and a second when subsequent alkoxy radical decomposition forms HO 2 and that HO 2 oxidizes NO to NO 2 followed by photolysis of that NO 2 . See Rosen et al. (2004) for a more complete discussion of these equations. The variable slopes observed in the plume are the result of the integrated instantaneous production rates. While there is no simple analytical expression for the integrated production, in what follows we use the instantaneous production rates to assess whether we can explain the changes in the slope from point to point throughout the plume. Taking the ratio of the two instantaneous production rates above we can find the relationship between the observed slope and the average branching ratio of the VOC mixture given by: where (ᾱ) is the average branching ratio. A slope of 60 O x / AN for example, implies (ᾱ)=3.2% while a slope of 20 implies (ᾱ)=9.1%. We can also calculate the production rate of O x and ANs based on known or assumed reaction rates and nitrate formation branching ratios as given by the Leeds Master Chemical Mechanism (MCM). Table 1 summarizes the values of α and k OH used in this analysis, the median concentrations of each VOC as observed within 5 h of Mexico City and the associated instantaneous production rates for O 3 and ANs. The calculated instantaneous O x / ANs slope for each point is the ratio of the total P (O x ) (column 5) to the total P ( ANs) (column 6). The calculated slope from Table 1 (for data within 5 h of Mexico City) is then 20 268/347, which is approximately 60 O x / AN. Parallel calculations are performed for each of the different ranges of plume age. In addition to the 30 hydrocarbon species measured on-board the DC8, we have estimated concentrations of another 12 which are marked in the table by bold italic font with underlining. The nine unmeasured alkenes were estimated based on correlations with 1-butene observed at the T1 site. The three longchain alkanes were estimated based on correlations with nheptane observed in Houston. Observations described in Parrish et al. (2009) and others show that emission ratios of related hydrocarbons (as inferred from observed hydrocarbon ratios) vary little from city to city even though emission rates vary dramatically. The estimated compounds add ∼5% to P (O 3 ) and 13% to P ( ANs). Note also that we have not included ANs themselves as precursor molecules in the initial calculations, as has been done previously (Rosen et al., 2004;Cleary et al., 2005). The specific molecular structure of the nitrates will determine both their OH reactivites and di-nitrate branching ratios and the extreme variation in possi-ble values for both of these parameters in the evolving plume makes them hard to approximate with a single value.

ANs sources in Mexico City
The largeᾱ (10.5%) implied by the 16.8 O x / ANs slope observed in and around Mexico City indicates a significant role for ANs in the photochemistry of the region. AN formation at such high (ᾱ) has a strong influence on O 3 formation rates because RONO 2 formation is competitive with HNO 3 formation as a chain termination reaction for HO x and NO x catalytic cycles. Given this large influence it would be valuable to understand the AN source molecules in detail.
There are large differences between the observed (17) and calculated (60) ANs v. O x slope near Mexico City, which implies the existence of unexplained photochemistry. This could be due to a combination of (1) the presence of unmeasured compounds with large nitrate branching ratios, (2) underestimates of currently accepted branching ratios for measured compounds, (3) a large component of di-nitrate formation from ANs (which would produce 2 molecules of NO 2 in the TD-LIF). Before further evaluation of these options it is useful to develop an independent check on the inferred α. One such independent check is to compare the observed ratio of (C 1 -C 5 nitrates)/ ANs to the ratio calculated from the instantaneous production rate given by Eq. (5). Namely: where α j and k OH+RHj are reported branching ratios and OH rate constants for only the hydrocarbon precursors to the C 1 -C 5 nitrates measured by GC and α i and k OH+RHi are reported or estimated branching ratios and OH rate constants for the entire suite of observed and estimated hydrocarbons given in Table 1. (C 1 -C 5 nitrates) are predicted to be 27% of ANs in Mexico City based on the relative production rates but they were only observed to be 10%. As with the discrepancy between calculated and observed ANs v. O x slopes, this indicates an underestimate of AN production. We should also note that, similar to the analysis performed above regarding the relative production of O x and ANs, we have assumed that the loss rates of both C 1 -C 5 nitrates and ANs are slow relative to production. The lifetimes of the simple C 1 -C 5 nitrates are well constrained by both laboratory and field measurements with typical photolysis lifetimes of ∼13 days (2pentyl nitrate) to 10 days (methyl nitrate) and lifetimes to oxidation by OH (at OH=3×10 6 ) of 2 days (2-pentyl nitrate) to ∼130 days (methyl nitrate). We calculate an overall loss rate of 2.7 ppt/h for C 1 -C 5 nitrates in Mexico City (as compared 7224 A. E. Perring et al.: The production and persistence of RONO 2 in the Mexico City plume to an instantaneous production rate of 31 ppt/h). At (15 h, 35 h) we calculate an instantaneous production rate of (2.7, 0.5) ppt/h and a loss rate of (0.65, 0.4) ppt/h. The lifetime of ANs as a class of compounds is more difficult to determine as the exact lifetime of any particular alkyl or multifunctional nitrate will depend on the specific structure. As a first guess we can make the approximation that, on average, they have an OH rate constant similar to that of toluene (a molecule with a moderately reactive double bond), that 20% of the reactions with OH lead to loss of nitrate functionality and release of NO 2 (as opposed to creating a more functionalized nitrate) and that the photolysis rate is similar to n-pentyl nitrate (one of the larger and shorter-lived nitrates for which we have a well-characterized photolysis lifetime). We then calculate a loss rate of 13.6 ppt/h in Mexico City, 1.6 ppt/h at 500 km and 1.3 ppt/h at 1000 km as compared to production rates of 638, 66 and 14 ppt/h at 0, 500 and 1000 km respectively.
The analyses described above examining the production rate of ANs relative to both O x and C 1 -C 5 nitrates lead to remarkably similar conclusions. As outlined previously, the slope (17) of the correlation between ANs and O x observed in Mexico City implies an average branching ratio for nitrate formation of 10.5%. In contrast, the average branching ratio for nitrate formation calculated from current chemical mechanisms is 3.2% and the O x / ANs correlation implies that the rate of ANs production is 3.3 times higher than expected. In corroboration, the observed fraction of ANs comprised of C 1 -C 5 nitrates is 10% rather than the expected 27% implying that ANs production is 2.7 times higher than expected. Below we examine some possible explanations for the higher-than-expected ANs production rate.
First we consider the possibility of unmeasured nitrate precursors. A large burden of long-chain (>C 10 ) alkanes has been proposed based on observations of SOA in large urban centers and analysis of diesel exhaust (Robinson et al., 2007). Since long-chain alkanes have high nitrate formation branching ratios (∼35%), adding them to our inventory could bring the observed and calculated branching ratios into agreement. A concentration of 0.5 ppb of a compound that reacted gaskinetically with OH and had a branching ratio of 35% would result in a calculated O x / ANs slope near Mexico City of 25 O x / AN (compared to the observed value of 17) and a calculated (C 1 -C 5 nitrate)/ ANs ratio of 8% (compared to the observed value of 10%).
A second possible explanation is that current estimates of nitrate branching ratios for some compounds that were measured are too low. Most notably, Mexico City is observed to have a large burden of aromatic compounds and there currently exists no aromatic oxidation experiment in the presence of NO x that has shown carbon closure. The nitrate branching ratios used here are those used in MCM which are estimated using an expression proposed by Carter and Atkinson (1989). The full implementation of the MCMv3.1 aromatic oxidation scheme is described by Bloss et al. (2005).
In addition to uncertainty in the yield of nitrates from the initial oxidation of aromatic compounds, there is a high degree of uncertainty in the yields of ring-opening vs. ring-retaining products (Hamilton et al., 2003;Jenkin et al., 2003;Wagner et al., 2003;Wyche et al., 2009). The products of ringopening pathways could themselves have high nitrate yields and significant nitrate formation could thus result from oxidation of the first-generation products of aromatic oxidation. A doubling of the nitrate branching ratio for all aromatic compounds would decrease the calculated ANs v. O x slope to 39 and the calculated (C 1 -C 5 nitrates)/ ANs ratio to 22% which is an improvement. While uncertainties in nitrate formation from aromatic compounds alone are unlikely to bring the calculations and observations into complete agreement, nitrate formation from aromatics and their oxidation products is significant in Mexico City and further study is warranted.
The third possibility is that ANs themselves are the unaccounted-for nitrate precursors. As noted above, they were not included in the initial calculation due to uncertainties in rate constants and branching ratios but it is highly likely that there is appreciable di-nitrate formation in the chemical environment of Mexico City. The calculations performed for datasets in Granite Bay and Houston assumed an OH rate constant of 1.6×10 −11 molecs/cm 3 /s and a branching ratio of 5%. If we include ANs in the calculation here using these parameters, the calculated ANs v. O x slope and (C 1 -C 5 nitrates)/ −ANs ratio are minimally changed. However, many of the 1st generation nitrates in Mexico City should be large molecules that retain at least one double bond. They may therefore react relatively quickly with OH and should have reasonably high branching ratios. If we assume that they have an OH rate constant of 7×10 −11 cm 3 /molecs/s, comparable to estimates of OH rate constants for isoprene nitrates (Perring et al., 2009a;Giacopelli et al., 2005) and a branching ratio of 17%, then the calculated ANs v. O x slope is 25 and the calculated (C 1 -C 5 nitrates)/ ANs ratio is 0.08 which is comparable agreement to that achieved by the addition of 0.5 ppb of unmeasured long-chain hydrocarbons.
We should also note that the nighttime reaction of NO 3 with alkenes is known to give rise to ANs production without associated O 3 production (Warneke et al., 2004). The nitrate formation rate from NO 3 -initiated oxidation is typically much higher than for OH-initiated oxidation and this reaction could represent a significant source of ANs. Based on concentrations of NO, NO 2 and O 3 observed at the T1 site in the hours before sunset (medians of 1 ppb, 8.6 ppb and 63 ppb respectively) and a nighttime temperature of 10 • C, we calculate a possible total combined NO 3 and N 2 O 5 (an NO 3 reservoir species) production of 1.4 ppb over the course of a typical night. Therefore, even if all of the available NO 3 were to react with alkenes with a 70% nitrate yield, it would lead to a maximum production of 1 ppb of ANs which is ∼20% of observed daytime concentrations and would perturb the O x / ANs slope by a similar amount, far smaller than the observed discrepancy. In addition, significant nighttime concentrations (100's of ppts) of NO were observed at the T1 site at night, presumably due to local NO emissions. NO emissions would inhibit the accumulation of appreciable concentrations of NO 3 by titration. If we instead calculate potential NO 3 and N 2 O 5 production based on median concentrations of NO, NO 2 and O 3 observed aboard the NASA DC-8 (0.4 ppb, 1.2 ppb and 68 ppb respectively), the potential production is much smaller (∼110 ppt). It is therefore likely that typical nighttime concentrations of NO 3 and N 2 O 5 are substantially lower than the possible 1.4 ppb calculated above and that ANs production from NO 3 oxidation of alkenes is insignificant compared to typical daytime production rates of 0.3-0.5 ppb/h. Downwind of Mexico City, the possible nighttime production of NO 3 and N 2 O 5 from observed NO 2 and O 3 is only a few ppt.
In summary, measured ratios of ANs to O x in Mexico City are high, corresponding to an average implied branching ratio of 7-10%, and are much larger than calculated based on observed and estimated hydrocarbons. We identify three poorly known possible candidates for reducing the modeledmeasured difference: (a) the presence of unmeasured, longchain compounds with high yields, (b) underestimates of nitrate yields from organic molecules and (c) higher rates of dinitrate formation than have previously been considered in other locations and show that all of them are consistent with the data. All three should be subject to further investigation.

Evolution of the O x vs. ANs correlation
The calculated and observed O x v. ANs can generally only be directly compared in the near-field of a source region where production outweighs all other factors. As the plume ages the most reactive primary VOC are depleted, enhancing secondary OVOC and resulting in a mixture that produces ozone and ANs in different ratios than the initial mixture does. These effects have been described previously for observations in Houston and Granite Bay where an increase in the slope of O x / ANs (a decrease in α) over the course of the day was interpreted as due to the increase in non-nitrate producing ozone precursors. The slope was higher in the afternoon because of increased concentrations of O 3 precursors such as CO and CH 2 O that are generated from the oxidation of other hydrocarbons but do not form nitrates upon oxidation. Similar processes are likely driving the variation in the slope of ANs vs. O x in the Mexico City plume as it evolves downwind. We do find that, although loss becomes more important over time, production outweighs loss for both O x and ANs over the entire range of photochemical ages sampled by the DC-8. As noted above, for ANs the loss rate is ∼2%, ∼2.5% and ∼10% of the instantaneous production rate at 0, 500 and 1000 km respectively. The instantaneous photochemical loss rate for O x is never more that 0.5% of the production rate anywhere in the plume. In what follows we use the observed evolution of the ANs vs. O x correlation and the (C 1 -C 5 nitrates)/ ANs ratio in conjunction with the calculated instantaneous production rates to asses the spatial extent to which the plume is impacted by the unknown chemistry identified in Sect. 4.1.
In order to compare the calculated instantaneous production rates to the observed (C 1 -C 5 nitrates)/ ANs ratio and ANs vs. O x correlation we must use a kind of stepwise integration. The instantaneous and observed slopes or ratios would only be expected to match exactly in the near-field of the source region over which the VOC mixture is invariant. As the plume ages and the VOC mixture changes, the observations reflect a linear combination of all of the previous instantaneous production rather than the single instantaneous slope calculated for a given point. For instance, in the case of the slope of the ANs vs. O x correlation, the observed slope after some number, n, of elapsed time intervals ( t) is given by: Clearly for the first time interval, given correct chemistry, the overall slope (M 1 ) should be similar to the instantaneous slope ( (O x ) 1 / ( ANs) 1 ). If the calculated and the observed slopes matched reasonably well for the first time interval we could simply calculate M 2 using the equation above and compare it to the observed M 2 . Since, however, the observed and calculated slopes are entirely dissimilar at point 1, the observed and calculated slopes at point 2 would be different regardless of whether the calculated instantaneous production was correct over t 2 because the second step in the calculation is impacted by the error in the first. What we really want to assess is whether the calculated instantaneous production rates during t 2 can explain the change between M 1obs and M 2obs . To do this we calculate M n for n>1 using the observed M n−1 slope as follows: Figure 7a shows the observed O x v. ANs slope as compared to that expected based on stepwise integration of the calculated instantaneous ANs and O x production rates performed as described above. The first point shown in Fig. 7a is a direct comparison of the calculated instantaneous O x / ANs with the observed O x / ANs correlation in Mexico City. We then calculate successive slopes by combining the observed slope at the earlier time with the instantaneous production rate at the later time. This prevents an error in the initial condition from propagating through the entire calculation. The calculated slope in Mexico City and just downwind (blue) and calculated (green) ∑(C 1 -C 5 nitrates)/∑ANs ratio as a function of photochemical age. Shaded areas, again, represent the 1σ variation of the data. The observed (blue) and calculated (green) (C 1 -C 5 nitrates)/ ANs ratio as a function of photochemical age. Shaded areas, again, represent the 1 σ variation of the data. is considerably higher than the observed slope but the agreement improves substantially by >30 h of photochemical aging. Again, we can corroborate this result using the ratio of (C 1 -C 5 nitrates)/ ANs (Fig. 7b). The observed and calculated (C 1 -C 5 nitrates)/ ANs ratios converge within 15 h. Thus, the compounds responsible for the excess ANs observed in Mexico City likely have short oxidative lifetimes and are consumed early in the plume evolution. Any of the above hypotheses would be consistent with this behavior.

Impacts of ANs chemistry on predicted O 3 production
The slope of O x / ANs was observed to be remarkably high in Mexico City as compared to other urban locations The formation of alkyl and multifunctional nitrates affects our ability to predict O 3 production both because it represents a direct alternative to the generation of NO 2 and thus O 3 from the RO 2 +NO reaction and because of the less direct feedback on RO 2 and HO 2 concentrations. ANs formation has not been evaluated in detail in the current generation of chemical models, a fact we suggest should be remedied in the near future. Any regional or global model, aimed at predicting O 3 concentrations, that fails to take this chemistry into account will over-predict O 3 production as a result. We can estimate the magnitude of this effect using observed radical species and a series of simple equations. The instantaneous gross O 3 production is defined as: HO x loss occurs through production of HNO 3 , ANs and organic peroxides (HOOH, ROOH or ROOR). The initiation reaction of OH+RH is not considered a HO x loss because, although it consumes one molecule of OH, it results in the almost instantaneous production of an RO 2 molecule. Production of HNO 3 or an alkyl nitrate consumes a single HO x molecule while production of hydrogen peroxide or organic peroxides consumes two HO x molecules so L(HO x ) can be written as: Where α is the nitrate branching ratio. Setting P (HO x ) = L(HO x ) and inserting measured values of OH, HO 2 , NO and NO 2 , we can solve the quadratic equation to find the RO 2 concentration. In what follows we have calculated RO 2 for each of the times for which we extracted an effective branching ratio from the observations in Sect. 3.3 above. Taking a water concentration of 0.6%, we use k OH+NO2 =1.22×10 −11 cm 3 /molecs/s, k HO2+HO2 = 2.74×10 −12 cm 3 /molecs/s, k HO2+RO2 = 8×10 −12 cm 3 /molecs/s and k RO2+RO2 = 6.8×10 −14 cm 3 /molecs/s based on JPL evaluation number 15 (Sander The absolute ozone production rate calculated using Eqs. (7) and (8) with no nitrate formation (blue), with nitrate formation using the branching ratio calculated using the MCM mechanism as shown in Table 1 (magenta) and with nitrate formation using the branching ratios inferred from observed slopes of ANs vs. O x in the Mexico City plume (green). Again, the shaded area represents the spread of the data from the 1 σ variation of the data. et al., 2006) assuming generic RO 2 behaves as C 2 H 5 O 2 . It should be noted that the calculated RO 2 concentration and therefore the calculated P (O 3 ) is sensitive to the assumed value of k RO2+RO2 . When OH+NO 2 slow sufficiently to permit accumulation of peroxy radicals, RO 2 is approximately equal to HO 2 and HO 2 +RO 2 is the most important of the three peroxide formation reactions. We use the median observed values of OH, HO 2 , NO and NO 2 and calculate RO 2 concentrations using both the inferred α's and α=0. We then use these RO 2 concentrations to calculate P (O 3 ). The top panel of Fig. 8 shows the percent change in calculated P (O 3 ) as a result of including the formation of ANs. Near Mexico City, where alkyl nitrate production is high, the effect of including this chemistry in a model is to reduce the calculated P (O 3 ) by 40%. The bottom panel of Fig. 8 shows P (O 3 ) using α=0 (blue squares and line), using α's as would be calculated using the MCM and observed hydrocar-bons (magenta squares and line) and using the α's inferred from the O x / ANs correlation (green squares and line). The effect persists, albeit minorly, downwind of the city and the integrated effect of neglecting alkyl nitrate formation leads to significant errors in the calculated O 3 production over the lifetime of the megacity plume.

Conclusions
ANs have been observed to be a significant (∼10%) fraction of NO y in the immediate vicinity and well downwind of Mexico City. Hundreds of ppts of ANs were observed over the Gulf of Mexico. The ANs fraction of NO y increases with increasing photochemical age and indicates continued production of ANs far from the source region. The source of ANs is larger in the Mexico City plume than in other urban locations. The effect of ANs on calculated P (O 3 ) is ∼30%, a value that is unexpectedly large and that implies careful consideration should be given to the role of AN formation in discussion of any O 3 control strategies. We find that these observations of high AN concentrations require one of more of the following: the presence of unmeasured, long-chain hydrocarbons with high nitrate formation rates, underestimates in nitrate formation rates for some fraction of the measured compounds, and/or a larger role for the formation of di-nitrates than has been considered previously. The evolution of the plume downwind indicates that the unknown compounds responsible are relatively short-lived.