Measurement of ambient aerosols in northern Mexico City by single particle mass spectrometry

Continuous ambient measurements with aerosol time-of-flight mass spectrometry (ATOFMS) were made in an industrial/residential section in the northern part of Mex- ico City as part of the Mexico City Metropolitan Area-2006 campaign (MCMA-2006). Results are presented for the pe- riod of 15-27 March 2006. The submicron size mode con- tained both fresh and aged biomass burning, aged organic carbon (OC) mixed with nitrate and sulfate, elemental car- bon (EC), nitrogen-organic carbon, industrial metal, and in- organic NaK inorganic particles. Overall, biomass burning and aged OC particle types comprised 40% and 31%, respec- tively, of the submicron mode. In contrast, the supermicron mode was dominated by inorganic NaK particle types (42%) which represented a mixture of dry lake bed dust and indus- trial NaK emissions mixed with soot. Additionally, alumi- nosilicate dust, transition metals, OC, and biomass burning contributed to the supermicron particles. Early morning pe- riods (2-6 a.m.) showed high fractions of inorganic particles from industrial sources in the northeast, composed of inter- nal mixtures of Pb, Zn, EC and Cl, representing up to 73% of the particles in the 0.2-3µm size range. A unique nitrogen- containing organic carbon (NOC) particle type, peaking in the early morning hours, was hypothesized to be amines from local industrial emissions based on the time series profile and back trajectory analysis. A strong dependence on wind speed and direction was observed in the single particle types that were present during different times of the day. The early morning (3:30-10 a.m.) showed the greatest contributions from industrial emissions. During mid to late mornings (7- 11 a.m.), weak northerly winds were observed along with the most highly aged particles. Stronger winds from the south picked up in the late morning (after 11 a.m.), resulting in a decrease in the concentrations of the major aged particle types and an increase in the number fraction of fresh biomass particles. The highest wind speeds were correlated with the highest number fraction of fresh biomass particles (up to 76% of the submicron number fraction) when winds were coming directly from fires that were located south and south- east of the city based on MODIS fire count data. This study provides a unique clock of hourly changes in single parti- cle mixing state and sources as a function of meteorology in Mexico City. These new findings indicate that biomass burning and industrial emissions can make significant contri- butions to primary particle loadings in Mexico City that are strongly coupled with local meteorology.


Introduction
The Mexico City Metropolitan Area (MCMA) is a megacity that allows for a unique opportunity to study air pollution. High levels of criteria pollutants are a product of the city's high population density, meteorology, and unique geographical location. Both gas and particle phase contaminants are generated that degrade human health and et al., 2005;Jimenez et al., 2004;Marr et al., 2006). Electron microscopy was used to infer the mixing state and transformation of soot particles (Johnson et al., 2005). Chemically resolved PM 2.5 mass distributions were obtained using a variety of techniques including Aerosol Mass Spectrometry (AMS), and other filter based techniques (Salcedo et al., 2006). Salcedo et al. (2006) was also able to track mass concentrations of select, non-refractory aerosol components with a higher time resolution than previously reported by Chow et al. (2002a), while showing a general agreement between the two studies. Newer source apportionment measurements during the MCMA-2003 campaign were able to classify major sources of particles including industrial emissions using factor analysis . Industrial emissions in Johnston et al. (2006) 15 were found to be well correlated with Na and Zn as well as other metals.
Many earlier studies of aerosol size and composition in Mexico City from 1990 to the present day were carried out as a part of major campaigns funded through an international effort (Molina, 2002). Prior to 2006, there were three major research initiatives that measured aerosol physico-chemical properties: The Mexico City Air Quality Re-

Sampling site -T0
The ATOFMS instrument was located at the Instituto Mexicano del Petroleo (IMP) in the northern part of Mexico City (19 • 29 ′ 23.60 N, 99 • 08 ′ 55.60 W). This was one of the 5 three supersites selected for the MILAGRO measurement campaign to characterize the transport of emissions from the urban areas in the MCMA to the surrounding regions. Figure 1 shows the geographical location of the IMP site, referred as T0 (urban site). Measurements were located in a secondary structure on top of a five-story building. Sampling lines were placed >10' above the structure's roof to minimize the effects of 10 sampling from the building ventilation exhaust ports. To the north was an 800 m high mountain, Cerro del Chiquihuite, that served to block most of the winds coming from the north. An industrialized area existed to the west, while urban areas resided to the east and south. A dry lake bed of Lake Texcoco was located to the east. A busy roadway was located on the east side of the site with traffic jams during most of the 15 day and street vendors cooking primarily during the morning and afternoon.

ATOFMS measurements and clustering analysis
The ATOFMS is an instrument designed to measure real-time size and chemical composition of aerosols. The specific model of the instrument used in Mexico City is described in (Gard et al., 1997). The aerosols are drawn through a nozzle inlet where the 20 gas undergoes a supersonic expansion, and the particles are accelerated to a specific terminal velocity depending on their aerodynamic size. The particle velocity is determined by measuring the time-of-flight between two 50 mW diode pumped, solid state, frequency doubled Nd:YAG lasers operating at 532 nm. The single-particle scattering EGU intensities from the two light scattering channels were acquired and saved along with the other single particle data as described in Moffet (2005) (Moffet andPrather, 2005). The particle size is then calculated from the speed using a calibration curve generated with known sizes of standard polystyrene spheres. The speed of the particle is also used to time the arrival of the particle in the ion source region of the dual-polarity 5 time-of-flight mass spectrometer. Once the particle is in the source region, a frequency quadrupled Nd:YAG laser operating at 266 nm with a typical pulse energy of 1.2 mJ desorbs and ionizes each particle. The ATOFMS measures both the positive and negative mass spectra of each particle simultaneously. The ATOFMS has wide dynamic range capabilities. This is accomplished by taking 10 the two signals from the two mass spectra being measured (positive and negative ion) and splitting them into an attenuated (30 dB A total of 1.6 million particles were sized and chemically analyzed with the mass spectrometer. The typical percentage of particles producing both size and chemical information was 50%; of these particles, 88% produced both positive and negative spec-20 tra. This percentage showed little variation over the study, indicating chemical matrix effects did not play a major role . Data from the ATOFMS were imported into a Matlab database program known as YAADA (http://www.yaada.org). Once in the Matlab database, the particles were split into four groups: sub and supermicron having wide dynamic range and non-wide dynamic range. The 48 000 particles 25 from the four subsets of particles were separately classified using ART-2a, a clustering algorithm (Song et al., 1999), run with a vigilance factor of 0.80 and a learning rate of 0.05. The clusters resulting from each analysis were matched to the rest of the particles in the complete dataset. In order to classify 1.4 million out of 1.6 million par-Introduction EGU ticles, 60 sub-micron and 200 super-micron clusters were considered (for each of the four groups) and accounted for 88% of the chemically analyzed particles. The 200 000 particles not classified made up a large number of sparsely populated clusters. The unclassified particles did not have any major temporal spikes indicating that the original number of particles given to ART-2a was sufficient. The particle clusters resulting from 5 the ART-2a analysis were grouped by hand into 15 general particle types. Hourly scaling functions were derived by scaling the ATOFMS data with size distribution data acquired with an aerodynamic particle sizer (APS, TSI, Inc.) using the method developed previously by our group . These scaled data were then used to derive mass concentrations of the specific particle classes by assuming 10 the densities suggested by .

Results and discussion
For the 3.5 weeks that the ATOFMS operated at the T0 site, unique anthropogenic particle types and mixing states were observed in northern Mexico City. These unique particle types primarily contained different metals and organic nitrogen species. The 15 majority of particles in the accumulation mode were either identified as biomass (from meat cooking or biomass burning), or organic carbon (OC). In the coarse (supermicron) mode, inorganic dust types were found to dominate. For all particle types, hourly time series (Sect. 3.4) and chemically resolved size distributions were obtained (Sect. 3.2). Average diurnal trends indicate that industrial emissions primarily occur in 20 the early morning and that biomass and OC make the largest contributions to aerosol mass during the early morning to late afternoon hours. To identify possible source regions, a concentration field analysis was performed by combining ATOFMS time series with stochastic Lagrangian back trajectories (Sect. 3.5). 3.1.1 Carbonaceous particle types OC: Organic carbon (OC) particles were one of the most abundant particle types observed during this study. These particles are similar to those detected in other ATOFMS field studies Noble and Prather, 1996;Pastor et al., 2003) and vehic-15 ular source characterization studies (Sodeman et al., 2005;Toner et al., 2006). OC particles are typified by a large number of hydrocarbon envelope peaks that start with the base carbon peak 12n C + n . In Fig. 2a, the major hydrocarbon-containing peaks are identified as 27 C 2 H + 3 and 43 C 2 H 3 O + . Many spectra also contain a large peak at m/z 39 (K+), indicating agglomeration between OC and biomass types. Biomass: Previous ATOFMS studies have provided a solid basis for the identification of biomass particles (Guazzotti et al., 2003;Silva et al., 1999). Particles classified as biomass always have 39 K + as one of the largest peaks in the positive ion mass spectrum. Additionally, the biomass mass spectrum contains carbon marker ions. Na + is more prominent in the biomass particles than it is in the other carbonaceous particle 25 types (with the exception of the NaEC type). Particles having these chemical signatures can be emitted as a result of biomass burning or cooking operations (Silva, 2000 EGU NaEC: The NaEC type is characterized by elemental carbon cluster ions at spacing of 12 m/z unitsin both the positive and negative ion spectra. This type has a dominant 23 Na + peak. This combination of EC and Na ions is consistent with observations in other single particle mass spectrometry measurements of heavy duty vehicle (HDV) emissions (Toner et al., 2006) and field studies (Guazzotti et al., 2001;Liu et al., 2003;5 Noble and Prather, 1996;Pastor et al., 2003). ECOC: The Elemental Carbon/Organic Carbon (ECOC) type is characterized by a positive ion mass spectrum that is dominated by clusters of carbon atoms. In addition to the major elemental carbon markers, minor signals occur from organic carbon envelopes along with many of the typical OC markers.

10
High Mass OC: The high mass organic carbon class was made to be a separate particle type due to the presence of hydrocarbon envelopes in the positive ions extending above 100Da. Typically these hydrocarbon envelopes have a ∆m/z = 14, which is due to successive losses of a 14 CH 2 group. Although not shown in the figure, the hydrocarbon envelopes can extend out to m/z = 200 and above. High mass negative ions 15 have been shown in lab secondary organic aerosol (SOA) studies (Gross et al., 2006) and have been attributed to oligomeric species. Positive ion laser desorption ionization mass spectra at 266 nm with such high mass signatures are typically attributed to polycyclic aromatic hydrocarbons (PAH) (Gross et al., 2000). Vanadium: Vanadium particles have been identified with other analytical methods as 20 well as single particle mass spectrometry in studies of light duty vehicle (LDV) emissions (Sodeman et al., 2005), from industrial urban areas (Noble and Prather, 1996;Tolocka et al., 2004), and from coal and oil fired power plant emissions (Suarez and Ondov, 2002). Particles containing vanadium produce a very unique mass spectrum with peaks at 51 V + and 67 VO + . Oxalate was seen in the negative mass spectrum at 25 m/z = 89, having the chemical formula 89 C 2 O 3 OH − . Secondary sulfate and primary vanadium are a result of emissions from fossil fuel combustion. Oxalate may be from either biomass burning, or VOC oxidation followed by subsequent cloud processing (Chebbi and Carlier, 1996;Morawska and Zhang, 2002). The fraction of the oxalic acid Introduction EGU mass formed by cloud processing is currently a topic of research (Ervens et al., 2004;Kanakidou et al., 2005). Nitrogen-containing organic (NOC): One of the new particle types detected in this study includes the NOC particle type. The "NOC" label is used due to the large peak at m/z =58 which we hypothesize is due to 58 C 2 H 5 NHCH + 2 which has been identified 5 by ATOFMS and in other laboratory studies (Angelino et al., 2001;Pitts et al., 1978). There is also a grouping of peaks at m/z =212-215 of unknown identity. Although the peaks are small, they occur on almost every NOC particle detected.
3.1. as the major peaks, but there is also significant signal from 79 PO − 3 and 35,37 Cl − . Coarse mode Na and K have been found to be in soil dust, biogenic material, and sea salt (Beddows et al., 2004).
AlSi: 27 Al + was frequently found to be mixed with 23 Na + , 39 K + and 35 Cl − . If Al was 15 associated with silicon markers 60 SiO − 2 and 76 SiO − 3 , the particle was placed in the Al/Si class. The presence of these peaks suggests that the particle was an aluminosilicate species common to mineral dust. 7 Li was another common peak seen in the positive spectrum of the Al/Si type. The peak at m/z =56 could occur from 56 Fe + or 56 KOH + ; and possibly even contributions from 56 CaO + when there is a corresponding peak at 20 m/z =40, indicating 40 Ca + . Ca: At m/z = 40 40 Ca + stands out as the largest peak in the positive ion spectrum accompanied by smaller peaks at m/z = 56 and 57 due to 56 CaO + and 57 CaOH + .
Previous studies have described the ability of CaCO 3 dust to react with nitric acid to form Ca(NO 3 ) 2 (Krueger et al., 2004;Laskin et al., 2005). It is likely that the calcium 25 dust shown herein has undergone this reaction as indicated by the negative ion mass spectrum which shows intense markers for nitrite and nitrate at 46

EGU
Minor peaks are seen in the positive spectrum at 24 and 27, likely due to Mg and Al, respectively.

Metal-rich particles
PbZn: Zn is one of the largest contributors to the positive ion spectrum occurring at m/z = +64, +66, +67 and +68. Often internally mixed with Zn, Pb shows up at m/z = 5 +206, +207, and +208. Other major peaks in the positive spectrum were 23 Na + and 39 K + . It has been observed in other studies that Pb, Zn and Na were associated with the industrial areas in northern Mexico City (Chow et al., 2002a;Flores et al., 1999;Johnson et al., 2006;Miranda et al., 1994). In addition to 46 NO − 2 and 62 NO − 3 , 35 Cl − was one of the most abundant markers in the negative ion spectrum. In general, zinc 10 and lead chlorides have relatively low boiling points (732 • C and 950 • C respectively), and their precursors may be present in high temperature combustion sources such as waste incinerators (Hu et al., 2003;Olmez et al., 1988;Ondov and Wexler, 1998). Upon cooling, these compounds will condense into the solid phase, forming submicron Cl-containing particles.

15
PbNa: It was common for Pb to occur without Zn, so the Pb/Na particles were separated to highlight this difference in mixing state. The Pb/Na type has Pb as the major transition metal marker. As for the PbZn type, Na was typically the largest peak in the positive ion mass spectrum, followed by potassium. Another similarity between Pb/Na and Pb/Zn is the appearance of the 35 Cl − peak in the negative mass spectrum. Introduction EGU 3.2 Chemically resolved particle size distributions Figure 3 presents the average mass distribution of the major particle types detected with the ATOFMS during the MCMA-2006. This size distribution was determined by scaling the ATOFMS data to the APS using a separate density for each particle class to transform the volume distribution into a mass distribution. In this study, we used 5 the density value of 1.9 g/cm 3 for carbonaceous particles, 2.7 g/cm 3 for dust particles, 1.9 g/cm 3 for Na-rich salt particles, 2.0 g/cm 3 for biomass emission particles and EC rich particles, and 1.9 g/cm 3 for the rest of the particle types. This follows the method developed by . By integrating the size distribution shown in Fig. 3, we determine the average PM 2.5 mass to be 43.5 µg/m 3 . This mass concentration is close to the value of 44.34 µg/m 3 obtained by Chow et al. (2002a, b) at the Xalostoc site (XAL in Fig. 1; close to T0), as well as that obtained at the RAMA site (38.1 µg/m 3 ) during the same time period as the current measurements. The RAMA site was located 1 km away from the T0 site.
With the exception of the high mass OC type, which peaked around 1 µm, carbona-15 ceous particle types are typically located in the accumulation mode between 0.1 and 1 µm, whereas the super-micron coarse mode particles were inorganic dust and salt particles as seen in previous ATOFMS studies (Noble and Prather, 1996). The high mass OC type may occur at larger sizes due to the higher degree of chemical aging or fog/cloud processing these particles have undergone. Given that this particle type 20 typically peaked at night when the relative humidity was high, this is likely the case . The lead and zinc types have interesting size characteristics: the PbNa particle type is primarily supermicron, whereas the PbZn particle type occurs at smaller sizes with a mode at about 850 nm. If these particles were formed by condensation, this size 25 characteristic may be due to different concentrations and physical properties of the vapor. Vanadium particles occur primarily in the submicron portion of the size distribution. Based on their size, the submicron metals are most likely a result of combustion 3.3 Analysis of mixing state using the peak search method In addition to the cluster analysis presented above, a separate analysis involving a peak search method was employed to highlight trends in the aerosol mixing state of the various particle types. A series of markers were selected to represent the different primary and secondary species. The m/z values used for primary and secondary species are shown in Tables 1 and 2. All ion intensities were set to be above or equal to an absolute peak area (≥100 units) that is just above the noise level (<50). To see which of the selected markers were associated with the various particle types, the particles with the species defined in Tables 1 and 2 were intersected with the major particle types derived from the ART-2a clustering analysis (Sect. 3.1). Figures 3 and 4 show the results of these intersections, where the color scale represents the fraction of particles of each major type (y-axis) associated with a particular marker (x-axis).

Mixing state of secondary species
The presence of secondary species on the various particle types (Fig. 4) provides an 15 indication of the type of chemical processing the particles have undergone in the atmosphere. The key question is: are there differences in the associations of the major secondary species (e.g. sulfate, nitrate, and ammonium) with each of the major particle types? For instance, the EC and NOC types contain the fewest particles with ammonium, nitrate and sulfate, indicating that they may be freshly emitted. As these particles 20 age through condensation and coagulation, they can accumulate other markers to become other particle types such as OC, ECOC and even biomass. Ammonium is seen to be constrained mainly to the submicron ECOC and OC types, whereas coarse mode particle types that contain NH + 4 are mainly limited to the AlSi and Cu types. For the AlSi type, the NH + 4 does not necessarily come from the atmosphere because it is common 25 to find NH + 4 in soils (Schlesinger and Hartley, 1992), and therefore may not be indicative EGU of aging in this particle type. Inorganic nitrogen species, such as NO 2 and NO 3 ,mostly occur on primary inorganic particles such as dust. However, more than 50% of the fine carbonaceous particles are associated with NO 2 and NO 3 , and show nitrate is often associated with fine mode particles. Differences in abundance are most likely due to preferential partitioning of the 5 NO 2 /NO 3 to the coarse mode particle types (size distribution presented in Sect. 3.2). This is because of the competition between nitrate and sulfate for surface area. Since sulfate is non-volatile, it remains on the smaller particles (limited by gas diffusion), whereas the more volatile ammonium nitrate can evaporate from the smaller particles (Kelvin effect) and condense on coarse mode particles (Bassett and Seinfeld, 1984).

10
The Pb/Na, Cu, and Pb/Zn particle types are more strongly associated with NO 2 /NO 3 compared to any other particle class. The Pb/Zn type is primarily a fine mode particle type, and it is unusual for particles in this size range to be so strongly associated with nitrate. Given that these industrial metal types are likely to be freshly emitted, this is an indicator that these particle classes are emitted with an air mass that contains large 15 amounts of NO that reacted to form particulate nitrate. It is likely the NOx displaced the Cl that was originally in the particles.
The organic carbon markers used to highlight mixing with other particle types were chosen to be 43 C 2 H 3 O + and C 2 O 3 OH − (oxalate). It is apparent that 43 C 2 H 3 O is mainly associated with the fine mode particles but occurs in over 40% of the supermicron 20 particles (with the exception of the PbNa type). This organic carbon likely comes from secondary reactions and gas-to-particle conversion. The oxalate marker is an indicator of the presence of oxalic acid. Oxalic acid may be emitted as a part of vehicle exhaust and biomass burning (Chebbi and Carlier, 1996;Falkovich et al., 2005;Kawamura and Kaplan, 1987) but is most commonly formed through secondary processes 25 including photochemistry followed by condensation, fog processing or aerosol surface reactions (Blando and Turpin, 2000;Faust, 1994;Kawamura and Ikushima, 1993;Yao et al., 2002). Oxalate is associated with biomass and vanadium types, as shown in Fig. 4, where 37 and 45% of the particles in these classes contain the oxalate marker,

Mixing state of primary species
A marker for EC, 36 C 3 , was chosen to highlight the distribution of carbon -especially in the non-carbonaceous particle types. This marker can come from secondary species as well, but is more commonly detected in primary particles. In Fig. 5, it is seen that the 5 PbZn and Cu types are mixed with EC. This result, combined with the predominately fine mode size distribution (Fig. 3), provides further evidence that the PbZn and Cu particles are products of high temperature combustion. While 36 C 3 can be used to identify particle classes that are internally mixed with EC, 40 Ca can be used to identify unique sources of EC. Source studies indicate that particles containing both Ca and 10 EC are primarily associated with heavy duty vehicle emissions (Toner et al., 2006). The mixing of elemental carbon with Ca is observed by searching for the typical EC markers together with Ca (CaEC). From Fig. 5, it is apparent that about 50% of the EC is associated with Ca, suggesting diesel vehicles produce a substantial fraction of EC/soot in this region.

15
Chlorine (m/z -35 and -37) is mainly associated with the inorganic particle types. This is supported by Fig. 5 which shows a strong degree of internal mixing between metals, K, Na and Cl. Cl is also strongly associated with the AlSi type and moderately associated with the NaK and Ca dust particle types. For Ca dust, there is much less Na and K associated with Cl compared to the other inorganic particle classes. A 20 secondary source of Cl is HCl, which can be formed by heterogeneous displacement when primary particles composed of NaCl, KCl, PbCl 2, or ZnCl 2 react with acidic (i.e. HNO 3 or H 2 SO 4 ) gases. The resulting HCl can then partition to other particles. This suggests that for the metals and AlSi dust, the Cl is primary and for the Ca dust, the Cl is secondary. It has been shown in previous studies that the mixing between Cl and 25 Ca is due mainly to secondary processing by HCl gas (Sullivan et al., 2007).
In Fig. 5, K particles are present in both the coarse and fine mode particles, whereas Na particles appear to be constrained mainly to the coarse mode. Caution is necessary 6427 Introduction  . This may explain why a large fraction of the organic carbon particles appear to contain K. On the other hand, it is likely that organic particles also contain K, given the high relative intensity of the K ion peak. Biomass particles are a prime example of a submicron particle type that definitely contains K and organic carbon 5 as an internal mixture. Comparison of the OC Art-2a clusters from Mexico City and Riverside, CA reveals that the peak at m/z=39 makes a larger contribution to Mexico City particles, indicating biomass represents a more significant source in Mexico City.

Temporal characteristics of single particle types
The temporal characteristics of the single particle types can provide a great deal of 10 insight into the origin of the particles. In this section, the temporal profiles for sub-and super-micron particle types are described in detail. Three major events are labeled on the temporal profiles shown in Fig. 6. The first event (E1) was a large dust event where the PM 10 concentration exceeded 1 mg/m 3 . The second event (E2) was a holiday weekend and the third event (E3) was a heavy rain event. Figure 7 shows an average 15 diurnal cycle of scaled mass concentrations representing the average of 19 days for the study.

Temporal profile of the entire study
Examining the study as a whole, major features in the overall submicron temporal trends (Fig. 6a)  EGU with industry. After the rain event (E3), the fractional contribution from the biomass decreased while OC, NOC, and metal particles increased. This trend continued for the rest of the measurement period, during which evening rain events occurred on a regular basis. Overall, the submicron classes exhibited a fairly strong diurnal behavior which will be examined further in Sect. 3.4.2.
5 Figure 6b shows the temporal profiles of the major supermicron particle types. There was one particularly large dust event caused by strong winds on 16 March 2006 beginning at 16:00 CST (E1) during which the measured PM 10 exceeded 1 mg/m 3 at the San Agustin and Xalostoc RAMA sites (near the dry lake bed Texcoco, SAG and XAL in Fig. 1) and almost 700 µg/m 3 at the La Villa RAMA site closest to T0. The wind blew 10 from these sites towards the west in the direction of T0. Our results show that the Na/K and Al/Si particle types dominated the chemical composition for this time period, thus indicating that the dry lake bed and fugitive sources nearby are major contributors to these particle types. This is consistent with the observations of Chow et al. (2002a and b), who found concentrations of Al, Si, K, and Na to be highest around the SAG and 15 XAL sites compared to other sites around the city. Figure 7 shows the average diurnal trends for the most abundant submicron particle classes. Metal and NOC particles reached their highest concentrations in the early morning hours between midnight and 10:00 a.m. The most dominant particle 20 types in the early morning (from 03:00-8:00 a.m.) were the OC and biomass particle classes. As the day continued, the relative proportion of ECOC and biomass increased. Biomass was the most abundant particle class from 08:00 a.m. to 05:30 p.m. Frequently, in the late afternoon, numerous fires were observed on the periphery of the city. In the late evening into the morning, the concentration of the OC particle types 25 increased compared to the other classes. This may be due to increased gas-to-particle phase partitioning of OC as the temperature decreased, or due to less dilution caused by a lower boundary layer. If the latter were true, this would indicate that more of the EGU OC particles were retained within the basin at night . Although it appears that biomass particles are the major source of particulate mass in the afternoons, one must use caution when interpreting this result. In the afternoon, there is more secondary organic carbon coating all of the particle classes -including the biomass class. Furthermore, coagulation can also cause a great deal of external mixing on a timescale of about 12 h (Jacobson, 2002). Because the ATOFMS has a high sensitivity to K, it is likely that if an OC particle coagulated with a biomass particle, the resulting particle would be classified as a biomass particle. Therefore, the ATOFMS would see more biomass and less OC particles in the late afternoon. On the other hand, the flow conditions in the afternoon were much different than in the morning and 10 the relative increase in biomass particles may be indicative of influences from different sources. This aspect will be discussed further in the following section using a back trajectory analysis.

Average diurnal trends
Elemental carbon showed a bimodal diurnal temporal profile with the largest temporal peak at 7:30 a.m. and the second mode occurring at 11:30 a.m. at the same time 15 as the total mass concentration peaked. According to Fig. 8, the EC particles sampled during the early morning peak had smaller aerodynamic diameters (number distribution mode =0.20 µm) than those sampled at the later peak (number distribution mode =0.38µm). In Fig. 8, it is also shown that the EC particles sampled later in the day had more ion intensity from NO − 3 , NH + 4 , C 3 H 3 O + , and K + . With the exception of K + , these 20 markers are indicative of secondary photochemical processing. Early morning EC particles had more intensity from 12n C − n clusters and sulfates in the negative ions as well as more intensity from Na + and Ca + in the positive ion mass spectrum. Most of these markers indicate primary species emitted with the elemental carbon particles. Based on these facts, it can be concluded that the early morning peak of EC is comprised 25 of more freshly emitted particles than the EC particles which peak later in the day. In addition, the fact that the EC particles sampled later in the day have more K + signal indicates that a different source of EC particles begins to have a larger influence as the day progresses, or that they have undergone coagulation with K containing particles.  (Seibert, 1994) was carried out using back-trajectories from the FLEXPART model (Stohl et al., 2005). To accomplish this, 100 stochastic particles are released from T0 every 2 h and tracked for 3 days. All the positions of the particles every hour are summed into a gridded field indicating the source region 5 of the air mass at each release time. These gridded fields, called Residence Time Analyses, are then multiplied by the ATOFMS normalized particle counts at the release site and summed over the entire campaign. Potential source regions are highlighted by normalizing with the sum of the unscaled residence time analysis. The method was used to analyze data from the MCMA-2003 field campaign (de Foy, 2006a). Analysis 10 of CO data showed that the method was able to correctly identify urban emissions and analysis of SO 2 data identified possible impacts of the Tula industrial complex. This analysis was performed for each of the fifteen particle types seen during MCMA-2006; the results of this analysis are shown in Figs. 8-10. For particle types with only a few sharp peaks, a back trajectory corresponding to peak concentrations is shown. 15 The wind fields for the trajectories were simulated using the MM5 model with modifications for the land surface (de Foy and Molina, 2006c). Additional modifications were made to the simulation procedure (de Foy et al., 2006b). These include finer and larger domains with 92×116, 91×145 and 97×97 cells at 27, 9 and 3 km resolution respectively and 41 sigma levels in the vertical, as well as corrections to the sea surface latent 20 heat fluxes.

Spatial distribution of industrial emissions
The T0 site is located in the heart of the industrial sector of the MCMA. As a result, a large fraction of the particles detected at the T0 site are expected to be from local industrial emissions. Typically, local point sources emit pollutants over short timescales. Introduction EGU type, this is verified by looking at the FLEXPART back trajectory shown in Fig. 9 for the largest NOC spike occurring at 09:00 a.m. on 15 March, 2006. This figure indicates that flow was stagnant during this peak in NOC emissions, and thus the source was most likely local. Similar flow conditions were observed during other periods of the study when the NOC particle type was most abundant.

5
From Fig. 7, one can see that the peak of PbNa and PbZn types occurred from midnight to about 10:00 (CST). The concentration field analysis in Fig. 9 indicates source regions north of T0 for PbZn, as well as for PbNa (not shown). Concentrations of Pb and Zn in northern Mexico City have been historically high compared to other regions of Mexico City (Chow et al., 2002b). There is evidence for transport of Na and 10 Zn particles from the northern to southern parts of the city, as Johnson et al. (2006) have shown. At the same site, Salcedo et al. (2006) noticed sharp spikes of particulate Cl in the early morning hours. In this study, we observe early morning spikes of chloride as well, and show that it is internally mixed with Pb, Zn, Na and K. Johnson et al. attributed the Na and Zn to an industrial source in their factor analysis, and Salcedo et 15 al. suggested that most of the Cl was present as NH 4 Cl. The results obtained herein show that all Na, Cl, and Zn are likely industrial and internally mixed within the same particles (Fig. 5). The speciation of Cl does not necessarily have to be exclusively NH 4 Cl because NaCl, KCl, ZnCl 2 or PbCl 2 are also possible. For all of these species, displacement of Cl by NO 3 explains the strong association of the metal particle classes 20 with NO 2 /NO 3 .
3.5.2 Spatial trends of carbonaceous particle types Carbonaceous particle types in Mexico City are expected to be produced by a variety of urban and regional sources, both primary and secondary. The concentration field analysis shown in Fig. 10 for the OC particle type shows that these particles originated 25 from all across the basin. The EC particle type shows more intensity close to T0 and in the northwest. Because the EC particle type represents a freshly emitted particle type and also because concentrations are highest when winds are weak and variable, the 6432 Introduction

EGU
Concentration Field Analysis does not indicate a preferred source region. EC particles coming from the north may be due to combustion operations or other industrial operations in the north. Biomass particles show a strong correlation with flow through the gap in the mountains to the southeast of the city -a region where large fires occurred. This type of flow usually occurs in the afternoon when the biomass particles peaked 5 and is associated with vertical advection and basin venting (de Foy et al., 2005;de Foy et al., 2006c). There is also slightly more source strength for the biomass particles near the hillsides of the basin. This may be due to the fact that there were frequently numerous small fires on these hills that were used to clear fields for the upcoming crop season. Lastly, vanadium particles are also correlated with the gap flow from the south 10 with a smaller signature from the north-northwest. Johnson et al. (2006) had found a single vanadium peak time period during the MCMA-2003 that was strongly correlated with SO 2 and back-trajectories from the Tula complex. During MCMA-2006, there were half a dozen peaks in the vanadium types in addition to substantial diurnal and dayto-day variations. Some of the peaks are associated with SO 2 and trajectories from 15 the north, while others are associated with weak and variable winds followed by the strong gap flow sweeping through the basin. At this time, it is unknown whether there are any large sources of vanadium particles in the MCMA. Some of the major sources that burn fuel-oil are located in the north, where Fig. 10 shows a mild influence from the vanadium particle type. 3.5.3 Spatial trends in coarse mode particle types The concentration field analysis from the coarse (super-micron) mode particle types is shown in Fig. 11. AlSi, Ca and Fe dust all had similar spatial footprints where the highest concentration events correlated well with the flow from the northwest near the Tula region. For the Ca type, this would be consistent with the fact that the cement 25 plants in that region emit dust with high concentrations of Ca (Vega et al., 2001). The Fe particle type had slightly more influence from the northwest than did the AlSi or Ca types. The NaK particle type was significantly different from any other coarse mode 6433 Introduction EGU particle type, showing more contributions from the east to northeast. The dry lake bed Texcoco is located in the east, and has been historically correlated with increased quantities of Na and K compared to the rest of Mexico City. Also, the NaK particle type was far more distributed spatially than the other coarse mode particle types. It is also possible that the NaK type contains contributions from vegetative detritus and 5 biomass burning, thus providing a possible explanation for its broad spatial distribution and stronger southerly signature.

Conclusions
ATOFMS observations during MCMA-2006 provide chemical mixing state measurements with high temporal and size resolution. A new nitrogen-containing organic (NOC) particle type was detected, and lead particles internally mixed with Zn and Cl were also frequently observed. Both of these particle types peaked in the morning hours and were likely the result of industrial emissions. Given the spatial and temporal characteristics of particles with the PbZnCl mixing state, these particles are directly associated with the source of early morning particulate Cl. Furthermore, these same metal par- 15 ticles are likely associated with the source of high Pb and Zn concentrations in the north as seen in previous investigations. These particles and their chemistry will be the subject of a forthcoming paper. The most abundant particle types seen in the MCMA include biomass and other carbonaceous types for the submicron size range, and dust and inorganic types for 20 the coarse (super-micron) mode. The mixing of these different particle types with secondary species was analyzed in further detail using a peak searching method. Organic carbon was found to be on almost 50% of the coarse mode dust types, while 45% of the biomass and 37% of the vanadium particles were associated with the oxalate ion. It was determined that 58% of the EC particles and 73% of the ECOC particles contained 25 sulfate. This indicates that these freshly emitted particles contain a significant amount of other inorganic components as a result of the source and of secondary chemical EGU processing. A temporal analysis of the different particle types gives insight into the possible sources and transformations of the particles. Distinct diurnal variations for the OC and biomass particle types were observed with the Biomass and OC present as the most abundant particle classes in the early morning, then after 10:00 CST, biomass 5 particles took over as the most concentrated carbonaceous particle class. This suggests that either transport and/or aging processes served to transform the chemical composition of the carbonaceous particles. After 18:00 CST, OC particles start to increase in concentration while biomass particles decrease. This diurnal variation has a potential to serve as a "clock" of the aging and transport characteristics of the aerosol 10 in the heart of the MCMA, thus providing a timescale by which measurements can be compared with models. These temporal profiles provided by the ATOFMS were used, together with meteorological modeling to provide source regions for the major particle types. Knowledge gained from this analysis will ultimately be combined with emissions inventories for the MCMA to give the most likely sources for each of the major particle 15 types. borne and D. Collins helped organize and consolidate the shipment of equipment into Mexico as well as providing APS data. T. Perez and R. Cepeda provided much needed and timely logistical support. R. Ramos graciously provided gas phase and PM data from the RAMA monitoring network. We thank NSF for funding under ATM-0511803 and ATM-0528227. We also thank DOE for funding under DE-FG02-0563980. EGU ZMCM, Nucl. Instrum. Methods Phys. Res., Sect. B, 109, 459-464, 1996a. Aldape, F., Flores, J. , Garcia, R., and Nelson, J. W.: PIXE analysis of atmospheric aerosols from a simultaneous three site sampling during the autumn of 1993 in Mexico City, Nucl. Instrum. Methods Phys. Res., Sect. B, 109, 502-505, 1996b. Angelino, S., Suess, D. T., and Prather, K. A.: Formation of aerosol particles from reactions 5 of secondary and tertiary alkylamines: Characterization by aerosol time-of-flight mass spectrometry, Environ. Sci. Technol., 35 (15) Chem. Phys., 6, 1249-1265, 2006a EGU during the SCOS97-NARSTO, Atmos. Environ., 37, S239-S258, 2003. Pitts, J. N., Grosjean, D., Vancauwenberghe, K., Schmid, J. P., and Fitz, D. R.: Photo-oxidation of aliphatic-amines under simulated atmospheric conditions -formation of nitrosamines, nitramines, amides, and photo-chemical oxidant, Environ. Sci. Technol., 12 (8) Fig. 11. Concentration field analysis for select coarse mode particle types. High nondimensional number (red) indicates possible source regions, low numbers (blue) indicate areas with low or zero emissions.