Contribution of Biomass Burning to Carbonaceous Aerosols in Mexico City during May 2013

During the springtime fire season, wildfires and agricultural burning represent a potentially large contribution to air quality degradation in the Mexico City Metropolitan Area (MCMA). PM10 filter samples were collected at six different stations in May 2013, the month with the maximum reported regional fire counts from 2002 to 2013. Two regimes were identified considering changes in predominant wind direction and precipitation patterns inside MCMA. The filter samples were analyzed for water-soluble organic carbon (WSOC) and the biomass burning tracers including levoglucosan (LEV) and water-soluble potassium (WSK). LEV concentrations correlated positively with ambient concentrations of PM2.5 and PM10 (R = 0.61 and R = 0.46, respectively). Strong correlations were also found between WSOC and LEV (R = 0.94) and between WSK and LEV (R = 0.75). PM2.5 accounted for 60% of the PM10 mass concentrations. Our speciated measurements accounted for 37% of the total PM10 mass concentration and ~60% of the PM2.5 mass concentrations; the missing mass was attributed to crustal material (soil or dust) and carbonaceous aerosols that were not segregated into the WSOC fraction. Average LEV/WSOC ratios ranged from 0.015 in the first, smokier and drier part of the month, to 0.006 during the rainier end of the measurement period. Using previously reported LEV/WSOC emissions ratios, the estimated biomass burning contributions to WSOC ranged from 7–23% assuming LEV is stable in the atmosphere, and 8–57% when accounting for LEV photochemical degradation in the atmosphere. Thus, our findings indicate that primary emissions from biomass burning sources represent significant contributions to ambient WSOC and PM in MCMA during the springtime fire season.


INTRODUCTION
The Mexico City Metropolitan Area (MCMA) is one of the most populated megacities in the Northern Hemisphere with a population of 20 million people (Instituto Nacional de Estadística y Geografía, 2012).The city lies in an elevated semi-enclosed air basin where the air quality is strongly influenced by geographic, anthropogenic, and biogenic factors.The high-altitude geographical location favors photochemical reactions (Fenn et al., 2002) and the abrupt topography makes thermal inversions common (Secretaría del Medio Ambiente del Distrito Federal, 2012a).MCMA is comprised of dense economic and industrial activities and a fleet of over 5 million vehicles (Secretaría del Medio Ambiente del Distrito Federal, 2013).Additionally, emissions from an active volcano, agricultural activities inside and outside the air basin, and evergreen needleleaf forests covering the surrounding mountain ranges, contribute to biogenic emissions as well as combustion emissions when wildfires strike the area.
The fire season in Mexico occurs from March through June each year, representing approximately 80% of the total annual area burned due to wildfires nationwide (Comisión Nacional Forestal, 2009).During March 2003 and2006, two large-scale field campaigns were conducted in Mexico City: MCMA-2003(Mexico City Metropolitan Area -2003) and the MILAGRO (Megacity Initiative: Local and Global Research Observations) Campaign, respectively.Both field campaigns revealed wildfires as a potentially important source of pollutants to MCMA, especially the MILAGRO campaign, which occurred during a very active fire period (Aiken et al., 2009;Crounse et al., 2009;Yokelson et al., 2011).Wildfires were estimated to contribute on average about 50 ± 30% to the aged fine particle mass in the outflow from MCMA (Yokelson et al., 2007) and ~10% to the fine particulate matter inside MCMA (Aiken et al., 2010;Querol et al., 2008).
For the present study we obtained aerosol samples from MCMA during May 2013, which was the month with the maximum satellite derived fire counts.The estimated PM 2.5 emissions for May 2013 exceeded by 40% the average monthly simulated PM 2.5 emissions for all fire seasons from 2002 to 2013 (Wiedinmyer et al., 2011), over a quadrangular region encompassing the MCMA defined by latitudes 15 to 23 degrees and longitudes -103 to -195 degrees.The collected aerosol samples were analyzed for biomass burning tracer species concentrations, which were used together with additional air quality monitoring data to develop estimates of the contributions of primary emissions from fires to aerosol concentrations in MCMA during this fire season.

Filter Sample Collection and Ambient Concentrations of CO, PM 10 and PM 2.5
The Air Quality Monitoring Network of Mexico City (AQMN-MC) collects ambient PM 10 samples using Hivolume air samplers (Andersen/GMW Model 1200) every 6 days.Samples are taken onto glass fiber filters for a 24hour period (sampling starts at 00:00 hours and finishes at 23:59 hours, local time).We obtained 29 samples from 6 stations for May 4, 10, 16, 22, and 28 (XAL station did not sample on May 4) (Table 1 and Fig. 1).The sampling characteristics for each day and station are presented in Appendix A (Table A.1). Additionally, we obtained reported PM 10 and PM 2.5 concentrations from filter measurements, which were calculated in the AQMN-MC laboratory by differential weighing under temperature and humidity controlled conditions (22 ± 2°C and 40 ± 5% relative humidity).
Moreover, hourly CO and particulate matter (PM 10 and PM 2.5 ) mass concentrations for each sampling day were obtained for the selected AQMN-MC stations (Table 1).CO measurements were made by a Teledyne-API model 300E with a gas filter correlation CO analyzer.Particulate matter was determined using a FH62C14 continuous ambient particulate monitor in Nezahualcoyotl station, and a 1405-DF TEOM™ continuous dichotomous ambient air monitor in the rest of the selected stations.

Filter Sample Analysis
Two 25 mm diameter punches of each PM 10 filter were extracted in 10 ml of deionized water (18.3 megaohms -Thermo Scientific Nanopure), sonicated with heat for 1 hour and 15 minutes and passed through a 0.2 µm PTFE syringe filter to remove possible glass fiber residues.The aqueous extracts were analyzed for 12 carbohydrates (including levoglucosan (LEV)) the same day they were extracted using high-performance anion-exchange chromatography with pulsed amperometric detection (HPAEC-PAD).A Dionex DX-500 series ion chromatrograph with a Dionex GP-50 pump and a Dionex ED-50 electrochemical detector operating in integrating amperometric mode using waveform A following the method of Sullivan et al. (2011).The limit of detection (LOD) calculated, considering air volumes for each sampler (Table A.1 in Appendix A), was 2.26 µg or 0.00118 ± 7.7 × 10 -5 µg m -3 (Table B.1 Appendix B).
Seven ions including water-soluble potassium (WSK + ) were also analyzed from the liquid extract during the same day of extraction using a Dionex DX-500 series ion chromatograph with a Dionex CD-20 conductivity detector, Dionex IP-20 isocratic pump and self-regenerating cation/ anion SRS-ULTRA suppressor.To separate the inorganic cations, a Dionex IonPac CS12A analytical column with a 20-mM methanesulfonic acid eluent at a flowrate of 0.5 mL min -1 was used (Sullivan et al., 2008).To separate the inorganic anions we used a Dionex IonPac AS14A anionexchange column with a carbonate/bicarbonate eluent at a flow rate of 1 mL min -1 .The LOD for the ions was 0.36 µg m -3 .Two blanks over the whole study were analyzed and the average ion concentrations measured on the filter blanks were subtracted from each filter samples (Table B.2 in Appendix B).
Finally, water-soluble organic carbon (WSOC) concentrations were measured using a Portable Total Organic Carbon Analyzer (Sievers Turbo).This instrument converts organic carbon in the sample to carbon dioxide using UV light and chemical oxidation by reaction with ammonium persulfate.The instrument was run in turbo mode.For the filter samples from May 4 and 10, 200 µL of the aqueous extract were diluted in 9 ml of deionized water, because the concentrations exceeded the upper LOD of the instrument.The rest of the sampling dates had lower concentrations of WSOC, and therefore 1 mL of the aqueous extract was diluted in 9 mL of deionized water.The WSOC LOD for this study, calculated based on the lowest measurable WSOC value (100 ppb) multiplied by the  volume extracted and divided by each sampler flow rate, was approximately 6.13 × 10 -7 µgC m -3 .

Regimes
Based on the differences in the number of satellite derived fire counts, mean precipitation rates, and air mass source regions, as well as the ambient concentrations of CO (ppm), PM 10 (µg m -3 ) and PM 2.5 (µg m -3 ), we divided the study period into two regimes (Fig. 2).Regime 1 (May 1-11) had no measured precipitation at any of the sites, was more strongly impacted by fire emissions, had back trajectories predominantly from the west (discussed in the next section), and had higher mean PM 2.5 concentrations (54 µg m -3 ) at all six sites in MCMA.In contrast, Regime 2 (May 12-28) had a mean precipitation rate of 4 mm day -1 , lower mean fire counts, air mass source regions predominantly from the north, and lower mean PM 2.5 concentrations (35 µg m -3 ).

Fire Counts
Two analyses were conducted to estimate the biomass burning sources that impacted the six sampling stations.The first analysis determined the history of air masses arriving at each sampling site over each 24-hour sampling period (ending at 0600 UTC).For each sampling date, hourly 24-hour backward trajectories from the center of the MCMA were calculated using the NOAA HYSPLIT model (Draxler and Rolph, 2013;Rolph, 2013).We ran the backward trajectories using two meteorological fields: first, we selected the North American Mesoscale Forecast System (NAM) from the National Centers for Environmental Prediction, which reports data at a 12 km horizontal resolution; then, we used the Global Data Assimilation System (GDAS), which reports data at a 1 degree horizontal resolution.Both meteorological fields returned consistent air mass pathways.To represent a well-mixed boundary layer, a 3500 MASL height was selected assuming a mean boundary layer height of 4000 MASL, which was estimated by analyzing the different heights of the lifting condensation level from soundings at 12Z and 00Z for each sampling date (Oolman, 2013).The vertical motion method selected was "model vertical velocity", which uses the vertical velocity field from the meteorological data.The center of the MCMA corresponded to the Merced station (Table 1).Air masses arrived at MCMA primarily from the west on May 4 and 10; the prevailing winds shifted mid-month such that air masses arrived primarily from the north for the rest of the sampling days.
The active fires that were assumed to have the potential to directly impact MCMA were the ones located within a 400 km radius from MCMA in one or more of the eight intercardinal directions (north, north east, east, south east, south, south west, west and north west).Output from the Fire INventory from NCAR (FINN) version 1 (Wiedinmyer et al., 2011) was used to geographically locate wildfires, prescribed burns and agricultural burning within this radius using the geographic information system ArcMap (Environmental Systems Research Institute, 2012).
Analysis of the spatial distributions of emitting fires (Table 2) showed that most of the fires occurred to the west and south west of MCMA on May 4 and 10.These two sampling dates also had the largest total area burned, and dominated the total CO and PM 2.5 emissions.For the   (2010) reported that 63% of the fires were located within a 60 km radius from MCMA during the MILAGRO campaign, whereas in May 2013 ~81% of the fires occurred within an annular area bounded by radii of 200 km and 400 km from the MCMA center (Fig. 1).

Gravimetric Mass Concentrations and Reconstructions
The estimated study-average PM 10 to PM 2.5 mass ratio was 1.68 (from a zero-intercept linear regression, Fig. 3).This ratio was calculated from reported gravimetric mass concentrations of 24 hour PM 10 and PM 2.5 filter samples during May 2013 for all available AQMN-MC stations.Our estimate was higher than the PM 10 to PM 2.5 mass ratio calculated from data reported by Querol et al. (2008) during the MILAGRO campaign at an urban site (T0) inside MCMA, where the average PM 10 to PM 2.5 mass ratio was 1.25.The higher PM 10 /PM 2.5 mass ratio we observed in our study might reflect that MILAGRO campaign had closer fires and thus higher nearby PM 2.5 emissions.The lower PM 10 to PM 2.5 mass ratio estimated during the MILAGRO campaign was closer to the source ratio of 1.09 reported by McMeeking et al. (2009) for biomass burning source filter samples obtained during chamber burns conducted during the Fire Lab at Missoula Experiments (FLAME) studies.This series of experiments was aimed at measuring chemical, physical and optical properties of biomass burning smoke, and obtaining source marker profiles for a wide range of North American and other fuels (Sullivan et al., 2008).
To reconstruct the dry PM 10 mass concentrations from the species measured in this study, we assumed that the measured sodium was present as sodium chloride, and that the conversion of WSOC to total organic carbon required a multiplication factor of 2 to account for associated oxygen, nitrogen, and other elements (Hand and Malm, 2006).We summed the remaining (non-Na + ) ionic mass concentrations to estimate mass concentrations of other ionic compounds.The average PM 10 characterized by our measurements in this study was 37% ± 2% of the PM 10 gravimetric measurements at the AQMN-MC sites at a 95% CI.The unexplained mass concentrations are likely to be insoluble crustal material (soil or ash) and insoluble carbon compounds that are not accounted for in the WSOC fraction.If it is assumed that all of the analyzed species were in the PM 2.5 fraction, then the resolved composition (WSOC and ionic species) accounted for ~60% of the PM 2.5 mass concentrations.

Biomass Burning Emissions
We obtained the FINN emissions of CO and PM 2.5 of each fire location over Mexico during May 2013.Coupling the fire location from FINN with the vegetation and land use types from Instituto Nacional de Estadística y Geografía (2005), we estimated the contributions of various fuel types to the total emissions from the 608 fires influencing MCMA (over a 400 km radius) during the 5 sampling days (Table 2).As a result, we found that 64% of the total area burned corresponded to wildfires, which contributed ~77% to the total emissions of CO and PM 2.5 .Wildfires that occurred in pine and oak forests and in deciduous and semi-deciduous forests were the source categories that contributed the most (~70%) to CO and PM 2.5 emissions.Agricultural burning emissions contributed ~23% of total biomass burning emissions of CO and PM 2.5 .Estimates of open waste burning emissions were also calculated; however, these accounted for less than 0.5% of the total emissions.

Biomass Burning Tracers
Table 3 presents the biomass burning tracer results for the PM 10 filter analyses discussed in this section (full carbohydrate and ion concentration results are presented in Appendix B); all concentration data are at local temperature and pressure.In general, the highest concentrations for all the carbohydrates were found on May 4 and 10, while most of the lowest concentrations were observed on May 16.As expected, LEV was detected in every PM 10 filter sample analyzed and had the highest ambient concentrations among all the analyzed carbohydrates for all sampling days and sites.The mean LEV concentration was 0.158 ± 0.046 µg m -3 at 95% confidence interval (CI) (0.070 ± 0.020 µgC m -3 at 95% CI) with a standard deviation of 0.146 µg m -3 (0.065 µgC m -3 ).Our results are consistent with the concentrations found by Stone et al. (2008) during the MILAGRO campaign in 2006, who reported an average concentration of 0.151 µg m -3 (0.067 µgC m -3 ) and a standard deviation of 0.136 µg m -3 (0.060 µgC m -3 ) for samples from an urban site (results reported at local temperature and pressure).
The average LEV concentration during Regime 1 was 0.335 ± 0.018 µg m -3 at 95% CI (0.149 ± 0.008 µgC m -3 at 95% CI) and for Regime 2 was 0.049 ± 0.005 µg m -3 at 95% CI (0.022 ± 0.002 µgC m -3 at 95% CI), representing a decrease from the first to second regime of about a factor of 7 in mass concentration.The highest LEV concentrations coincided with the highest fire count day (May 4).Furthermore, the daily mean LEV concentrations were positively correlated with fire counts for all sampling dates (R 2 = 0.66), and with other FINN modeled products such as area burned (R 2 = 0.66), CO (R 2 = 0.59), and PM 2.5 (R 2 = 0.61) emissions rates (Fig. 4).

Correlations of Ambient Concentrations of CO, PM 10 , and PM 2.5 with LEV
Our results showed low coefficients of determination of CO with LEV ambient concentrations for all sampling dates and at all AQMN-MC sites (R 2 ranged from 0.08 to 0.50), with an overall correlation of R 2 = 0.06.We found a CO concentration decrease of 9% from Regime 1 to Regime 2, with a mean CO ambient concentration of 1.1 ppm for the second.The generalized low correlation and small differences Stronger correlations were found between particulate matter (PM 10 and PM 2.5 ) and LEV concentrations.PM 10 and LEV coefficients of determination for each site ranged between R 2 = 0.55 and R 2 = 0.89, with an overall R 2 of 0.42.On average, PM 10 concentrations during Regime 1 were 55% higher than during Regime 2. The mean PM 10 concentration for Regime 1 was 104 ± 4 µg m -3 at 95% CI and 67 ± 6 µg m -3 at 95% CI for the second regime.We expect PM 10 to be influenced by local sources of primary particle emissions, such as dust from paved and unpaved roads (Secretaría del Medio Ambiente del Distrito Federal, 2012b), while biomass burning primary emissions are mostly in the PM 2.5 fraction.Despite the measured LEV concentrations being from the PM 10 fraction of the aerosol, the strongest correlations were found between ambient PM 2.5 and LEV, with coefficients of determination for each AQMN-MC site ranging between R 2 = 0.69 and R 2 = 0.78 and with an overall R 2 of 0.62.For Regime 1, the mean PM 2.5 concentration was 59 ± 3 µg m -3 at 95% CI and for Regime 2 was 35 ± 2 µg m -3 at 95% CI.

Correlations of WSK + and WSOC with LEV
The measured concentrations of WSK + were higher than LEV concentrations (Table 3) (mean concentration of 0.50 ± 0.63 µg m -3 at 95% CI) and were positively correlated with LEV (R 2 = 0.75) in all the samples (Fig. 5(a)).WSK + concentrations in Regime 1 were twice as high as those found during Regime 2, with an average concentration in Regime 1 of 0.71 ± 0.02 µg m -3 at 95% CI, compared to 0.36 ± 0.04 µg m -3 at 95% CI during Regime 2. Our results show stronger correlations between WSK + and LEV than found in recent studies of ambient samples during fire periods in other countries (Brazil: R 2 = 0.38 reported by Urban et al. (2012) and no linear correlation found by Schkolnik et al. (2005); China: no linear correlation found by Cheng et al. (2013)).The higher correlation found in this study might be due to soil impacts collected on the PM 10 filters.
The WSOC concentrations also showed a strong correlation (R 2 = 0.94) with LEV concentrations among all the AQMN-MC sites.Also, the highest WSOC concentrations were found during the first regime with a mean concentration of 10.10 ± 0.41 µgC m -3 at 95% CI, and the lowest during the second regime with a mean concentration of 3.72 ± 0.26 µgC m -3 at 95% CI (Fig. 5(b)).
The mean LEV/WSOC (µgC µgC -1 ) ratios for each sampling date are shown in Fig. 6.The uniform distribution of this ratio over MCMA in most of the sampling dates suggests one regional source of LEV.Our LEV/WSOC ratio during Regime 1 (0.015 µgC µgC -1 ) is similar to a calculated LEV/WSOC ratio from averaged LEV and WSOC ambient concentrations reported by Stone et al. (2008) for a non-urban site outside MCMA during the MILAGRO campaign (0.017 µgC µgC -1 ).The difference between these periods of intense fire activity might be due in part to the site location, since all of our sampling sites were located in residential and industrial areas (Table 1), and due in part to the proximity of the sampling sites to the fires during MILAGRO, particularly if LEV is subject to degradation during transport in the atmosphere as has been suggested (Hoffmann et al., 2009;Hennigan et al., 2010).

Estimates of Biomass Burning Contributions to WSOC
To estimate the portion of WSOC observed in MCMA PM 10 samples that could be attributed to biomass burning primary emissions, we used the calculated LEV/WSOC ratios from this study (LEV/WSOC study ) for each Regime and reported LEV/WSOC ratios fom FLAME (LEV/WSOC FLAME ) (Sullivan et al., 2014) for needles (LEV/WSOC = 0.064 µgC µgC -1 ) as our low estimate, and leaves (LEV/WSOC = 0.095 µgC µgC -1 ) as our high estimate to represent the range of ratios relevant to the most important vegetation burned during our study.We then estimated the percentage of WSOC attributable to biomass burning sources (WSOC BB ) using Eq. ( 1).
The estimated biomass burning contribution to WSOC for each sample ranged from 7% to 23%.As expected, the first regime had a higher estimated biomass burning contribution (average 19%), more than twice that computed for Regime 2 (average 8%).Using the LEV and WSOC concentrations reported by Stone et al. (2008) for a non-urban site within the MCMA air basin during the MILAGRO campaign, we estimated the range of WSOC BB in that campaign as 10 to 36%.The similar but slightly higher contribution during the MILAGRO campaign might be related to the proximity of the active fires to MCMA (Aiken et al., 2010), as mentioned previously.
These estimated WSOC BB percentages were calculated assuming that LEV is inert in the atmosphere.However, recent studies have shown that LEV can degrade in the atmosphere in both clear air and in-cloud conditions (Hoffmann et al., 2009;Hennigan et al., 2010).Therefore, our previous calculations of WSOC BB contributions might underestimate the influence of biomass burning.
To account for LEV degradation, we used an average LEV lifetime of 1.1 days (the time for its decay to 1/e of the initial concentration) as estimated by Hennigan et al. (2010) as the maximum LEV degradation rate.For the minimum LEV degradation rate, we used a LEV lifetime of 5 days, estimated from the work of Hoffmann et al. (2009).Assuming a pseudo first order reaction for the rate of decay, where the lifetime is equal to the inverse of the reaction rate (τ = k -1 ), we calculated the initial LEV concentrations (µg m -3 ) at the point of emission (Eq.( 2)), assuming our measurements represent the final concentrations after the respective lifetimes, and choosing the time t as the point when the backtrajectory passed over the active fires.
After correcting the LEV concentrations for this atmospheric decay, we recomputed WSOC BB for the revised LEV/WSOC concentrations (Table 4).The revised estimates increased the biomass burning direct emission contributions to 8-57%.However, this finding is specific to the source profile (LEV/WSOC ratios from FLAME) used in the apportionment.Source profiles appropriate for the regions in Mexico in which fires occur in the springtime are needed to increase confidence in this estimate.

CONCLUSIONS
May 2013 represented a particularly active fire month in Mexico, with emissions from biomass burning having the potential to affect the air quality of the entire region in which the MCMA is located.We estimated that wildfires, mostly in pine, oak and deciduous forests, contributed 77% of the total fire emissions of CO and PM 2.5 .Results from laboratory analysis of PM 10 filters collected throughout the MCMA basin showed that species associated with biomass burning emissions, namely LEV, WSK + and WSOC, had mass concentrations in ambient air that were about 7, 2 and 3 times higher respectively, during Regime 1 as compared with Regime 2, attributable to a shift in precipitation patterns, predominant air mass pathways, and an overall decrease in fire activity as the month progressed.Correlations found between LEV concentrations and modeled biomass burning emissions suggested that the whole air basin was affected by transported emissions from fires occurring in the surrounding region.Correlations between LEV concentrations and ambient concentrations of PM 2.5 (R 2 = 0.62) also suggested a strong biomass burning influence.Likewise, a strong correlation (R 2 = 0.75) was observed between LEV and WSK + concentrations on the PM 10 filters that were analyzed.For both WSOC concentrations and LEV/WSOC (µgC µgC -1 ) ratios, the highest values were found during Regime 1.
PM 2.5 accounted for 60% of the PM 10 mass concentration.On average our laboratory characterization accounted for 37% of the total PM 10 mass concentration, suggesting large contributions to PM 10 mass concentrations from non-watersoluble aerosol components.Future work is needed to measure more constituents in order to explain the possible sources of the remaining mass.
Assuming that LEV is stable in the atmosphere during transport, 7-23% of the total WSOC was attributed to primary biomass burning emissions, depending on the source profile selected, with the highest percentage contributions during Regime 1.When the possibility of LEV degradation in the atmosphere during transport was considered, at least 8% and as much as 57% of the WSOC measured during Regime 1 was apportioned to primary emissions from biomass burning sources.Thus, we conclude that biomass burning sources had a significant to large impact on WSOC and PM 2.5 during May 2013.
The results from this study contribute to the limited number of estimates of the relative contributions of primary emissions from biomass burning to carbonaceous aerosols in MCMA during an active fire season.One limitation of this work is the lack of smoke marker source profiles that are specific to MCMA.Sampling closer to active fires, for different vegetation and different burn phases, is thus needed, and/or laboratory studies using fuels from the surrounding regions.Another limitation to estimating total impacts of biomass burning is that SOA formation from biomass burning sources is highly likely.To date there are no specific molecular markers for smoke-derived SOA, but work in this arena is ongoing, and other methods including modeling of the air quality in MCMA could be used to estimate the fire contributions to SOA.Finally, the FINN and MCMA databases may be missing some local fire sources, such as burning to clear the verges of roads and highways.Since these sources occur within MCMA they may have strong impacts on air quality, despite being much smaller in emissions magnitude than large wildfires.Given the importance of understanding sources contributing to MCMA air quality degradation in order to develop sound and effective mitigation strategies, further work is needed to estimate differences between sources on urban and nonurban sites, and biomass burning contributions to total carbon and ambient concentrations of particulate matter in MCMA.Such studies will ultimately be used to improve the emission inventories that are commonly used by decision makers to design air quality policies and emission source controls.
Fig. 1. a) PM 10 monitoring stations from AQMN-MC where 24-hour samples were taken during May 2013.Station locations from Dirección de Monitoreo Atmosférico (2013).b) Fire locations for Regime 1 are shown as dots and for Regime 2 as X symbols.The outer radius represents the 400 km radius considered in this study.The inner circle represents the 60 km radius from MILAGRO campaign.The thick line to the left is shown as an example of a typical HYSPLIT backward trajectory for Regime 1 and the one to the right as an example for Regime 2.

Fig. 2 .
Fig. 2. Daily average precipitation rates (Red de estaciones pluviométricas de la Ciudad de México, 2013) and PM 2.5 daily mean ambient concentrations (Dirección de Monitoreo Atmosférico, 2013) during May 2013.Data from rain gauges nearest to the AQMN-MC stations are reported.Dates within red boxes correspond to PM 10 filter sampling dates.No precipitation was reported over the last 5 days of the month.

Fig. 3 .
Fig. 3. Comparison of gravimetric mass concentrations of PM 10 and PM 2.5 for all available AQMN-MC stations in May 2013.Solid line is the linear regression of PM 10 mass onto PM 2.5 mass forced through the origin.Dashed line is the 1:1 line.Gravimetric mass concentration data from Dirección de Monitoreo Atmosférico (2013).

Fig. 4 .
Fig. 4. Correlation between mean LEV (µgC m -3 ) concentrations inside MCMA and daily totals of FINN products over a 400 km radius from MCMA for each sampling day (averaged over all the sampling sites).

Table 2 .
Fire characteristics over relevant spatial area surrounding MCMA for each sampling day.Only PM 2.5 estimates are reported in the FINN emissions dataset.Fire counts dataset from FINN.
May 16, 22 and 28 sampling dates, air masses arrived mostly from the north, and fire counts were reduced overall in the region, suggesting a lower fire impact for these sampling days.Compared to the MILAGRO campaign, the fires in May 2013 were farther away from MCMA.Aiken et al.

Table 4 .
Hoffmann et al. (2009)of biomass burning contribution to WSOC (WSOC BB ) for each regime (left section).Corrected LEV concentrations and LEV/WSOC ratios were calculated considering 24-hour LEV degradation rates.Upper LEV degradation rate fromHennigan et al. (2010).Lower LEV degradation rate based onHoffmann et al. (2009).The corrected percentages of WSOC BB calculated using reported FLA.