Emissions from biomass burning in the Yucatan

In March 2006 two instrumented aircraft made the ﬁrst detailed ﬁeld measurements of biomass burning (BB) emissions in the Northern Hemisphere tropics as part of the MILAGRO project. The aircraft were the National Center for Atmospheric Research C-130 and a University of Montana/US Forest Service Twin Otter. The initial emissions 5 of up to 49 trace gas or particle species were measured from 20 deforestation and crop residue ﬁres on the Yucatan peninsula. This included two trace gases useful as indicators of BB (HCN and acetonitrile) and several rarely, or never before, measured species: OH, peroxyacetic acid, propanoic acid, hydrogen peroxide, methane sulfonic acid, and sulfuric acid. Crop residue ﬁres emitted more organic acids and ammonia 10 than deforestation ﬁres, but the emissions from the main ﬁre types were otherwise fairly similar. The Yucatan ﬁres emitted unusually high amounts of SO 2 and particle chloride, likely due to a strong marine inﬂuence on this peninsula. As smoke from one ﬁre aged, the ratio ∆ O 3 / ∆ CO increased to ∼ 15% in < ∼ 1 h similar to the fast net production of O 3 in BB plumes observed earlier in Africa. The rapid change in O 3 occurs at a ﬁner 15 spatial scale than is employed in global models and is also faster than predicted by micro-scale models. Fast increases in PAN, H 2 O 2 , and two organic acids were also observed. The amount of secondary organic acid is larger than the amount of known precursors. Rapid secondary formation of organic and inorganic aerosol was observed with the ratio ∆ PM 2 . 5 / ∆ CO more than doubling in ∼ 1 . 4 ± 0 . 7 h. The OH measurements 20 revealed high initial OH levels > 1 × 10 7 molecules / cm 3 . Thus, more research is needed to understand critical post emission processes for the second-largest trace gas source on Earth. It is estimated that ∼ 44 Tg of biomass burned in the Yucatan in presents the ﬁrst detailed measurements of the initial trace gas and par- ticle emissions from ﬁres in the NH tropics (up to 49 species on 20 ﬁres). It includes the ﬁrst in-situ measurement of OH in a BB plume and measurements of numerous post-emission changes in trace gas and particle species in one plume. Only a few observations of the chemical evolution of BB smoke have they are only partially reproduced these measurements of smoke evolution this We estimate


Introduction
The MILAGRO (Megacity Initiative Local and Global Research Observations) project was designed to study the local to global atmospheric affects of pollution from megacities (http://www.eol.ucar.edu/projects/milagro). Megacities have a population >10 million and are rapidly increasing in number on the five most populated continents. 5 The first MILAGRO field campaigns occurred in March 2006 and studied the impact of trace gases and particles generated in Mexico City (MC, North America's largest metropolitan area) on regional atmospheric chemistry (Fast et al., 2007;Molina et al., 2007). Regional biomass burning (BB) was also studied because it is a major emission source in Mexico that peaks during the February-May dry season. Earlier papers esti-10 mated that BB located adjacent to MC accounted for ∼20−30% of the CO (and several other important trace gases) and about one-half of the particle mass in the March 2006 MC outflow (Yokelson et al., 2007a;J. Crounse private communication). Another very important source of BB emissions in the MILAGRO study region is the Yucatan, which accounts for 7% of Mexico's land area, but almost 30% of the total biomass burned 15 in Mexico annually (2002-2006 average) and almost 40% of the biomass burned in Mexico in 2006. (Section 4 describes the model used to generate these estimates.) From the perspective of the MILAGRO campaign, Yucatan BB emissions are important because MC and the Yucatan impact nearly the same regional environment and Yucatan emissions can be transported to MC or interact with the MC plume downwind of Introduction  (Ridley et al., 2004). Formaldehyde was measured by the NCAR difference frequency generation (DFG) airborne spectrometer (Weibring et al., 2007). The absorption (530 nm), scattering (550 nm), number, and size distribution of dry particles was measured at 1-10 s resolution by a particle soot absorption photometer (PSAP), nephelometer (TSI 3563), and optical particle counter (OPC) all deployed by the Uni-5 versity of Hawaii (Clarke et al., 2004). The total 550 nm scattering was converted to STP scattering and then PM 2.5 (µg/s m 3 ) using a dry mass scattering efficiency (MSE) of 4.8±1.0 obtained for a USFS TSI model 3563 during the gravimetric calibration carried out for the Twin Otter nephelometer. The absorption was used directly with the scattering to calculate single scattering albedo (SSA) or converted to an estimated 10 black carbon in µg/s m 3 using a mass absorption efficiency (MAE) of 12±4 (Martins et al., 1998). The NCAR Scanning Actinic Flux Spectroradiometers measured 25 J values at 10 s resolution (Shetter and Müller, 1999). An NCAR selected ion chemical ionization mass spectrometer (SICIMS) measured the hydroxyl radical (OH), sulfuric acid (H 2 SO 4 ), and methane sulfonic acid (MSA) at 30 s time resolution (Mauldin et al.,15 2003).

Printer-friendly Version
Interactive Discussion 2008) measured the organic aerosol mass (OA); the OA to organic carbon (OC) mass ratio; and non-refractory (NR) sulfate, nitrate, ammonium, and chloride (µg/s m 3 1 atm, 273 K) for the last 6 s of each 12 s measurement cycle. (The first 6 s of each cycle measured size distributions.) At times continuous 4 s particle chemistry averages were recorded instead. A proton transfer mass spectrometer (PTR-MS) measured CH 3 OH,5 acetonitrile (CH 3 CN), acetaldehyde, acetone, methyl ethyl ketone, methyl propanal, hydroxyacetone plus methyl acetate, benzene, and 13 other species in a 35 s cycle (Karl et al., 2007(Karl et al., , 2008).

Generalized airborne sampling protocol
The Twin Otter and C-130 were based in Veracruz with other MILAGRO research air-10 craft (http://mirage-mex.acd.ucar.edu/). The main goal of the Twin Otter flights was to sample fires and the C-130 also sampled a few fires. On both aircraft, background air (i.e. ambient boundary layer (BL) air not in plumes) was characterized when not sampling BB plumes. The continuous instruments operated in real time in background air. The discrete instruments acquired numerous spot measurements in background air. 15 These spot measurements should be representative since the continuous instruments showed that the background air was well-mixed on the spatial scale corresponding to the discrete sampling intervals.
To measure the initial emissions from the fires, the aircraft usually sampled smoke less than several minutes old by penetrating the column of smoke 150-600 m above 20 the active flame front. A few "fresh" smoke samples up to 10-30 min old were acquired at elevations up to 1700 m. The continuous instruments monitored their species while penetrating the plume up to five times per fire. On the Twin Otter, the AFTIR, MPS-3, and WAS were used for spot measurements in the smoke plumes. To allow calculation of excess concentrations in the smoke plume; paired background spot measurements 25 were made just outside the plume. The discrete instruments on the C-130 always acquired at least one sample, and usually several, that were within the fresh smoke plumes. The background for those samples was taken as the average of the discrete 775 Introduction Conclusions References Tables  Figures   Back  Close Full Screen / Esc

Printer-friendly Version
Interactive Discussion measurements in the background air near the plume.
More than a few kilometers downwind from the source, smoke samples are already "photochemically aged" and better for probing post-emission chemistry than estimating initial emissions (Goode et al., 2000;Hobbs et al., 2003). Both aircraft acquired some samples in aged plumes up to ∼14 km downwind and ∼1.5 h old (see Sect. 3.4). 5

Data processing and synthesis
Grab, or discrete, samples of both a plume and the adjacent background can be used to calculate excess mixing ratios (∆X, the mixing ratio of species "X" in the plume minus the mixing ratio of "X" in the background air). ∆X reflect the degree of dilution of the plume and the instrument response time. Thus, a useful, derived quantity is the 10 normalized excess mixing ratio (NEMR) where ∆X is divided by the "simultaneously" measured excess mixing ratio of another species (∆Y); usually a fairly long-lived plume "tracer" such as ∆CO or ∆CO 2 . The uncertainty in the NEMR includes a contribution due to differences in response times if 2 instruments are involved. A measurement of ∆X/∆Y in a plume up to a few minutes old is a molar emission ratio (ER). We computed 15 fire-average molar ER for each individual fire from grab or discrete samples as follows. First, if there is only one sample of a fire then the calculation is trivial and equivalent to the definition of ∆X/∆Y given above. For multiple grab (or discrete) samples of a fire, the fire-average, ER was obtained from the slope of the least-squares line (with the intercept forced to zero) in a plot of one set of excess mixing ratios versus another. 20 This method is justified in detail by Yokelson et al. (1999). When the AFTIR and the USFS WAS measured the same pair of compounds on the same fire, their data were combined in the plots as shown in Fig. 1a.
Emission ratios can also be obtained from the continuous instruments by comparing the integrals of ∆X and ∆Y as the aircraft passes through a nascent smoke plume. Introduction

Conclusions
References Tables  Figures   Back  Close Full Screen / Esc

Printer-friendly Version
Interactive Discussion through the plume of a single fire with continuous instruments (e.g. PM and CO 2 on the Twin Otter), we plot the integrals versus each other and obtain the ER from the slope; analogous to grab sample plots. For the C-130 we usually compared integrals for various species to the integrals for CO. The exception was the PTR-MS. For the PTR-MS we obtained ER to methanol averaged over the three C-130 fires by comparing the 5 integrated excess for all 3 fires to the integrated excess amount for methanol. Finally, it is sometimes possible to use a "proxy" to generate continuous data from discrete samples. For example, the ratio of HCN to NO y should not vary much throughout an individual plume. Thus, an estimate of the real-time variation in HCN can be obtained by multiplying the continuous NO y data by the HCN/NO y ratio measured intermittently in the plume. Subsequently, the "continuous" HCN trace can be integrated and compared to integrals from other continuous instruments such as CO. For two species measured discretely by the Caltech CIMS (SO 2 and HCN) it was possible to compare the "integralbased" ER to CO to the "plot-based" ER to CO. Compared to the integral-based ER, the plot-based approach returned individual fire ER that were identical when a few 15 minutes were spent in plume and up to 57% different for brief sample periods with an average difference of ±21%. It is important to note, however, that the ER obtained from the plot-based method did not show significant bias when averaged over several fires. Specifically, the plot-based ER was within 7% of the integral based ER averaged over 2 fires for SO 2 , and within 5% averaged over 3 fires for HCN. We take this as a rough 20 estimate of the additional uncertainty affecting the study-average ER calculated from discrete samples. The lack of bias makes sense since e.g. a slower-responding CO instrument could read a little low when entering the plume, but a little high when leaving the plume. Figure 1 illustrates typical analyte levels encountered and gives examples of ER derivations.
excess mixing ratios tend to vary slower in time and space making timing differences less critical. Section 3.4 discusses uncertainty in the aging plume in detail.

Estimation of fire-average initial emission factors
For any carbonaceous fuel, a set of ER to CO 2 for the other major carbon emissions (i.e. CO, CH 4 , a suite of non-methane organic compounds (NMOC), particle carbon, 15 etc) can be used to calculate emission factors (EF, g compound emitted/kg dry fuel burned) for all the emissions quantified from the source using the carbon mass-balance method (Yokelson et al., 1996). In this project, EF were calculated for all the individual fires sampled by the Twin Otter from AFTIR measurements of CO 2 , CO, CH 4 , and NMOC and WAS measurements of CO 2 , CO, CH 4 , and non-methane hydrocarbons Introduction  (Andreae and Merlet, 2001). We also assumed all the fires burned in fuels that were 50% C by mass on a dry weight basis (Susott et al., 1996), but the actual fuel carbon percentage could vary by ±10% (2σ) of our nominal value. EF scale linearly with the assumed fuel carbon fraction. Because much of the NO is quickly converted to NO 2 after emission, we also report an EF for "NO x as NO." For any species "X" we abbreviate 5 the EF as EFX. The last C-130 sample of nascent smoke (from Fire #3 on 23 March) was at 1700 m and thus about 10-30 min old. Immediately after this sample, the aircraft stayed in the plume and followed the smoke in a curving path for about 14 km down wind probing smoke that had presumably aged an additional ∼1.5±0.7 h relative to the sample of nascent smoke (14:16:48-14:18:31 LT) (Sect. 3.4). During the nominal aging interval 5 from 1.27-1.47 h the RH exceeded 100% indicating possible cloud processing. By 14:19:41 LT, the C-130 had climbed back above the boundary layer and set course for Veracruz.  the fires sampled (from 13:37-14:41 LT), and the hotspot locations (from Terra 1 at 11:16 LT and Aqua 1 at 12:46 LT). The fire, fire type, and hotspot matches are also in Table 1. In contrast to the 22 March flight when most of the fires found were DF, on 29 March all but one of the fires found were originally CR. Thus, of the six fires found, there was one DF, 3 CR, and 2 mixed (CR that escaped and also burned some woodland). 15

Fire-average initial emission factors measured on the Twin Otter
The fire-average emission factors (EF g/kg) measured on the Twin Otter are shown in Table 2 along with the study average EF for DF and CR fires. If molar ER are preferred for an application, they can be obtained from the EF in Table 2  MCE and demonstrates that much of the large variation in EFCH 3 OH (factor of ∼4.5) is correlated with the different MCE that occur naturally on biomass fires. In principle, it would be advantageous to incorporate this variability into emissions estimates. Unfortunately, at this time, MCE cannot be measured from satellites nor can seasonal trends in MCE be confidently assigned (Yokelson et al., 2008). For N and S species 5 EF variability also arises from the variable N and S components of the fuel (Yokelson et al., 1996(Yokelson et al., , 2008. We note that the sum of the EF for gas-phase non-methane organic compounds (NMOC) in Table 2 is 12.8 and 12.0 g/kg for CR and DF fires, respectively. However, oxygenated volatile organic compounds (OVOC) normally dominate the gas-phase NMOC emitted by biomass fires and in this study many OVOC 10 common in biomass smoke were not measured on the Twin Otter. In an earlier study of fire emissions with enhanced detection of gas-phase OVOC, the combined instrumentation could not identify about 50% of the gas-phase NMOC by mass (Yokelson et al., 2008). Therefore, we speculate that 30 g/kg is a conservative estimate of the real sum of gas-phase NMOC from these Yucatan fire types (see also Sect. 3.2).   (and Sect. 3.2) that the Twin Otter total NMOC is too low. The present study does, however, greatly expand the extent of speciation of Yucatan fire emissions. We compare our DF data to the most recent and extensive measurements of DF in the Amazon, a study which included both airborne measurements of nascent emissions and ground-based measurements of the smoldering emissions that cannot be sampled 5 from an aircraft (Yokelson et al., 2008). A few main features stand out. The largest difference occurs for initial emissions of particles. The EFPM 2.5 for the Twin Otter Yucatan DF is only 4.5±1.6 g/kg compared to 14.8±3.4 g/kg for the Amazon. Both of these values are significantly different from the recommendation of Andrea and Merlet (2001) of 9.1±1.5 g/kg, although their recommendation agrees remarkably well with 10 the average of the Yucatan and Amazon values (9.7 g/kg). A main factor contributing to the observed difference in EFPM 2.5 is probably that the March 2006 early dry season fires that we were able to sample in the Yucatan were much smaller and less "intense" than the late dry season fires sampled in Brazil. As discussed in detail by Yokelson et al. (2007b) and references therein, it is likely that larger more intense fires have 15 much larger particle emission factors. Since a large fraction of annual biomass burning occurs late in the dry season, the late dry season EFPM may better represent the annual particle production. Also, we show in Sect. 3.4 that the ∆PM 2.5 /∆CO ratio could increase by a factor of ∼2.6 in ∼1.4 h after emission. This factor times our initial EFPM 2.5 of 4.5 g/kg suggests that shortly after emission, about 12 g/kg of PM 2.5 have 20 been produced even by the small Yucatan fires considered here.
The Yucatan fires had higher mean EFCO 2 ; 1656±38 (1601 Amazon) and lower mean EFCO; 83±14 (108 Amazon). This is indicative of relatively more flaming combustion, which is also reflected in the higher MCE; 0.927 (0.904 Amazon). The higher MCE and a possible tendency for the biomass to be higher in nitrogen content may ex-Introduction HCOOH (below detection limit in Yucatan, 0.57 Amazon). The EF measured for a single understory fire in a tropical dry forest in Africa were all within the range of EF we measured for the Yucatan DF, except that the EF for total PM for the African understory fire was 13 g/kg (Sinha et al., 2004). There are not enough measurements of understory fires to determine if their EF are significantly 10 different from the EF for DF. For species not measured in this study, the EF measured for deforestation fires in Brazil are probably the best estimate for these, and global tropical forest fires (Yokelson et al., 2008). By extension, the EF values measured in this work, but not in Brazil (Sect. 3.2) are likely the best estimate for global tropical forest fires. 15

Comparison of Yucatan deforestation and crop residue fires
We also compare the EF from the two main types of fires we observed in the Yucatan. Despite the fuel differences, the EF overlap within one standard deviation for most species. However, the tendency towards higher emissions of organic acids and ammonia from crop residue fires is clear; CH 3 COOH (4.8 CR, 2.9 DF); HCOOH (2.7 CR, 20 below detection limit DF); and NH 3 (1.38 CR, 0.775 DF). "Higher than normal" emissions of these species were also observed from burning rice straw by  in a lab study.
The Yucatan DF emitted more HCHO than the CR fires we sampled (below detection limit CR, 2.6 DF). However, the rice straw fire sampled by   Interactive Discussion may not be very sensitive to a shift in the relative amount of these two main fire types for most species.

Fire initial emissions measured on the C-130
No CO 2 data was collected on the C-130 so we present ER to CO in Table 3. Also the fuels were not positively identified so the C-130 data was not included in the "fire-type 5 specific" comparison above. However, the extensive instrumentation on the C-130, and the well-designed flight path, provided a large amount of valuable data. The initial fire emissions of several species were measured for the first time in the field by the Caltech CIMS. These include H 2 O 2 , peroxyacetic acid, and propanoic acid. All these reactive compounds were present in significant amounts in the youngest 10 smoke samples. The presence of the peroxide species in the nascent smoke may reflect some fast initial photochemistry (e.g. recombination of peroxy radicals (RO 2 , HO 2 )), but there is no increase in the peroxide species with altitude in our samples. They are important as a HO x reservoir and the H 2 O 2 plays an important role (along with HCHO) in the oxidation of sulfur in clouds (Finlayson-Pitts and Pitts, 2000). 15 The NCAR SICIMS detected traces of H 2 SO 4 (∆H 2 SO 4 /∆CO, 5.4×10 −7 ±5.2×10 −7 ) and MSA (∆MSA/∆CO∼8.4×10 −8 ±1.3×10 −7 ) in the young fire emissions also for the first time. The initial amount of these species varies greatly probably due to variations in fuel S and plume reactivity. The combination of PTR-MS and the Caltech CIMS made it possible to detect many 20 gas-phase NMOC that have been measured from fires before, but not on the Twin Otter in this study. In Table 4 where the results from the two aircraft are combined (as detailed in Sect. 3.3) the sum of identified gas-phase NMOC is 22 g/kg. This value is close to the sum of identified gas-phase NMOC for Brazilian deforestation fires (25.8 g/kg) reported by Yokelson et al. (2008). However, both of these values are only ∼50% of the 25 true sum of gas-phase NMOC, which includes detected, but unidentified species that are mostly high molecular weight OVOC (Yokelson et al., 2008). The sum of the molar ratios to CO ( Interactive Discussion of the molar ER to CO for measurable NMHC was 0.0393. Thus, the OVOC/NMHC ratio in the initial gas-phase emissions was about 3.3 -or OVOC accounted for 77% of measured, emitted NMOC on a molar basis. The dominance of NMOC by OVOC in BB plumes causes significant photochemistry differences compared to fossil fuel burning plumes, where NMOC are dominated by NMHC (Singh et al 1995;Mason et al., 2001).

5
An important species detected by PTR-MS was acetonitrile, which is thought to be produced almost exclusively by biomass burning (de Gouw et al., 2003) and thus has value as a BB tracer with relatively long (few months) atmospheric lifetime. The ER to CO for this species averaged over the 3 C-130 fires was 0.0043. The recommended acetonitrile to CO ER for 18 Brazilian DF was 0.0026±0.0007 (Yokelson et al., 2008).
Without better information on the variability of this ratio in the Yucatan we cannot say it is different from the ER for Brazilian DF. HCN is another compound produced mainly by biomass burning and used as a tracer (Li et al., 2000). The ∆HCN/∆CO ER measured on the C-130 was 0.0108±0.008 (n=3) and the Twin Otter ∆HCN/∆CO ER for all fires was 0.0032±0.0016 (n=17). The combined average for both aircraft is about 15 0.007±0.006, which is not significantly higher than the ∆HCN/∆CO ER obtained for Brazilian DF (0.0063 ± 0.0054). The ∆CH 3 CN/∆HCN molar ER is fairly consistent for several recent field studies: 0.39 MC-area (J. Crounse private communication), 0.41 Brazil DF (Yokelson et al., 2008) and ∼0.26 in the Yucatan if we only consider the fires where both species were measured. The ∆CH 3 CN/∆HCN molar ER for laboratory 20 fires in six different tropical fuels was 0.56±0.31 .
As noted in Sect. 3.1.1, the Yucatan DF emitted more NO x than Brazilian DF, which could suggest a higher fuel N content in the Yucatan and the hypothesis that NO x is produced by oxidation of most of the N-containing compounds in the fuel. The fact that HCN and acetonitrile emissions are highly variable in both locations, but not larger 25 (on average) from the Yucatan fires suggests that these species may arise from the pyrolysis of only specific, N-containing precursor compounds in the fuel that vary in concentration, but on average are about the same in Brazil as in the Yucatan. In any case, potential, total fuel N differences do not provide a simple rationale for all our  (Yokelson et al., 2008). Both the Caltech CIMS and NOAA UV instruments confirmed higher than average emission of SO 2 from these fires. The average SO 2 /CO ER from the CIMS instrument reflects one more fire than sampled by the NOAA UV instrument. The CIMS average molar ER of 0.0173 is approximately 7 times larger than the average for tropical forest 5 fires (0.0024) quoted by Andreae and Merlet (2001). The reason for this could involve the fuel S content, which could be high in the Yucatan because of soil S; manure used as fertilizer; or deposition of marine sulfur. There are some very large SO 2 sources in central Mexico including volcanoes and petrochemical refineries (de Foy et al., 2007;Grutter et al., 2008). These sources are normally downwind of the Yucatan, but there 10 may be occasional meteorological circumstances where they contribute to sulfate deposition in the Yucatan.
The mean PM 2.5 mass ratio to CO (0.0684±0.0054, Table 3) obtained by coupling the UH nephelometer with the NCAR CO for the 3 C-130 fires was very similar to the average ratio obtained for all 17 fires on the Twin Otter (0.073±0.026). This suggests 15 that the two aircraft sampled a similar mix of biomass burning and confirms that the PM emissions were below the literature average for tropical forest fires. The UH PSAP allowed a rough measurement of black carbon (BC) based on light absorption. Martins et al. (1998) (2003) showed that EFBC increase with MCE (BC is a flaming combustion product) while CO correlates with smoldering combustion, which also tends to emit a "white" reflective smoke high in organic mass (Yokelson et al., 1997). Fires 1 and 3 sampled by the C-130 had updraft cores that directly rose to high altitudes in the boundary layer Interactive Discussion due to vigorous flaming, while Fire #2 had a lower buoyancy plume (C-130 flight video, http://data.eol.ucar.edu/). The plume dynamics are consistent with the BC and SSA data as Fires 1 and 3 had low SSA (0.67 and 0.73) and high BC/CO mass ratios (0.066 and 0.081) compared to Fire 2 (SSA 0.84, BC/CO 0.043). From the OPC data, assuming spherical particles, we calculated the dry surface area 5 of particles 0.15-3 microns in diameter, which should account for most of the particle surface emitted ( Table 3). The average value was 48±6.1 m 2 /g PM 2.5 . The lowest individual value was obtained for Fire #3 (41.1 m 2 /g PM 2.5 ), which was sampled at the highest altitude (1700 m) and likely reflected fast initial coagulation. The more aged upper haze layers had values near 30 m 2 /g PM 2.5 , and this is likely a good estimate 10 of the dry surface area after most of the coagulation is complete. Ambient RH in the upper boundary layer ranged from 70-100% so the ambient particles would have been significantly larger due to addition of water. Measurements of the growth of BB particles as they hydrate typically show an exponential increase in total scattering to about a factor of 2 near 80% RH, which is usually the cut-off for the measurement (Magi and 15 Hobbs, 2003). Thus, the particle surface area could certainly double as "dry smoke" becomes "wet smoke." Assuming an aged smoke layer dry PM 2.5 of 100 µg/s m 3 we obtain a dry particle surface area concentration of ∼3×10 −3 m 2 /s m 3 . Multiplying by a typical number of active sites per m 2 (10 19 , Bertram et al., 2001) suggests a "few ppbv" of active surface sites are available for heterogeneous chemistry in a typical wet 20 or dry smoke plume. This is a significant available surface area, but much smaller than the droplet surface area within clouds (up to 0.5 m 2 /s m 3 ). The tendency of smokeimpacted clouds to have more, but smaller droplets (Kaufman and Nakajima, 1993) can cause the surface area in smoky clouds to be 2-7 times larger than in clean clouds. Depending on the extent of smoke-cloud interaction this could be the most important 25 influence of BB on available surface area. The AMS coupled with the VUV-CO provided the mass ratio to CO for a suite of particle constituents: NR sulfate, nitrate, ammonium, chloride, and organic mass (Tables 3 and 4). Coupling the AMS with the UH nephelometer allowed an estimate of the  (Tables 3-5). These ratios can be compared to measurements in plume penetrations of nascent smoke from Brazilian fires by Ferek et al. (1998). (Since these ratios change rapidly after emission, it is best to compare initial ratios from similarly aged, very fresh samples.) Since Ferek et al., report a higher average ratio of PM4 to CO (0.10) this could bias a comparison of ratios 5 to CO between the two studies. Thus we compare our mass fractions of total PM 2.5 to the mass fractions of total PM 4 that they obtained for the average of four fire types shown in their Table 3. The average MCE for their data treated in this manner is 0.924; close to the average MCE (0.929) measured on the Twin Otter for all Yucatan biomass burning. 10 Fire #2, which may have been a forest fire (based on the video and photos) had the highest OA and lowest BC and is the C-130 fire that is most like the Brazil fires (Table 5). Fires 1 and 3 have much higher initial NR chloride, nitrate, and ammonium than Fire 2 or the Brazilian fires. This could indicate that these fires were burning crop residue and that the emissions were impacted by fertilizers. However, the high average EFCl − in 15 the Yucatan could also reflect wet deposition of marine aerosol on the fuels (McKenzie et al., 1996) "refractory" in other studies may be detected by the AMS. Finally, one African tropical dry forest fire emitted "total" PM that was only 23% OC (Sinha et al., 2004). The sum of the species analyzed on the C-130 accounted for 87±13% of the total PM 2.5 as calculated from our MSE-based approach. Other methods of computing total PM 2.5 on the C-130 could be used such as the size distribution coupled with the particle number 5 and an assumed density. But the agreement obtained from the MSE-based approach is adequate as each technique only claims accuracy of ∼25%. The main aerosol species not quantified on the C-130 is potassium (K). The K + signal in the AMS can reflect both surface and electron impact ionization making it difficult to quantify the amount of K in ambient particles. K was about 3.7% of particle mass in Ferek et al., but its incorporation into particles depends strongly on the amount of flaming combustion (Ward and Hardy et al., 1991) as may also be the case for chloride.

Overall combined initial emissions from Yucatan biomass burning
In this study we simply take the average of all the Twin Otter data and weight it equally to the average of all the C-130 data. As discussed in Sect. 4, this approach may weight 15 the emissions from CR fires too heavily to be a true regional average, but it allows us to use the valid measurements from mixed or unknown fire types. Since, we have also presented the emissions data in the original stratified form (Tables 2 and 3), this allows alternate coupling schemes for future applications. In any case, as noted above, the fire type may not affect the smoke chemistry dramatically except for organic acids, NH 3 , 20 and some PM species. In addition, the smoke transported away from the Yucatan will not have the same composition as the initial emissions due to rapid photochemistry detailed in Sect. 3.4. To couple data from both aircraft in our estimate we proceed as follows. Both molar ER to CO and EF are useful for regional estimates and only the former was measured 25 on the C-130. Thus, in Table 4 fires sampled by the C-130. In Column 6 the combined average ER to CO is shown for all species. When a species was measured on both aircraft or by two instruments we took the straight average of the two values as our preliminary estimate of the study average regional value (except for SO 2 as detailed above). Finally, in Column 7, the study-average initial ER from Column 6 have been converted to EF using the Twin Otter 5 average CO/CO 2 and the carbon mass balance method (Sect. 2.4.1). It is worth noting briefly that, on average, similar fires were sampled by each aircraft. The PM 2.5 /CO ratios are close as noted above. A best calculation of the methanol to CO ratio measured on the C-130 (allowing for sampling rate differences) gives an average value for the three C-130 fires of 0.0254, which is within 10% of the average measured by AFTIR for the 17 fires sampled on the Twin Otter (0.0282). The ∆CH 3 COOH/∆CO and ∆HCHO/∆CO ER measured on the two different aircraft were also very similar (Table 4). The means were not as close for NO x and HCN, but the standard deviation about the mean was large for both species on both aircraft. We do note that, the ∆HCN/∆CO ER measured on Fire #3 by the AFTIR on 22 March was 15 0.0037, which is not far from the value measured by the Caltech CIMS (0.0047) on the continuation of that fire the next day, though perhaps burning partly different fuels.

Photochemical aging of smoke (first 1.5 h)
Post-emission chemistry determines much of the atmospheric impact of smoke from fires. In this study, the C-130 first sampled Fire #3 at 1700 m where the smoke 20 would have been ∼10−30 min old and then immediately followed the plume downwind. About 14 km downwind, ∆CO in the plume suddenly decreased from values that were ≥10−20 times the variation in the background (∼5 ppbv CO) to only 2-3 times the background variation. The measurements continued beyond this point, but we do not discuss them since the excess values are highly uncertain. The average windspeed 25 measured in the aging plume was 9.6±4.2 km/h. Assuming the average winds were the same when the downwind smoke was originally emitted implies that the downwind smoke had aged up to an additional ∼1.5±0.7 h. In Fig. 4a, ∆O 3 /∆CO is plotted versus the estimated change in smoke age. A rapid increase in this ratio to ∼15% occurs in <1 h. Figure 4a also shows ∆O x /∆CO versus time where "O x " approximates the total odd oxygen. In this work the sum of O 3 , NO 2 , and PANs account for nearly all the odd oxygen. The rise in ∆O x /∆CO is very similar to that in ∆O 3 /∆CO confirming that O 3 is being produced through photochemical 5 oxidation of NMOC (Crutzen et al., 1999). Yokelson et al. (2003) measured a rise in ∆O 3 /∆CO to ∼9% in ∼0.7 h on 3 isolated BB plumes in Africa and the ∆O 3 /∆CO observed in this work after ∼0.7 h of smoke aging (∼8%) is close to the value observed in the African plumes. The chemical evolution of one of the above-mentioned African plumes was measured in great detail  and Trentmann et al. (2005) constructed a comprehensive photochemical model for comparison with those measurements. The model agreed with the measured rate of increase in ∆O 3 /∆CO only if plausible, but unconfirmed, heterogeneous reactions were added; or if the measured initial emissions of NMOC were increased by 30% (on a molar basis) as a surrogate for unmeasured NMOC. The latter assumption is consistent with our earlier statement 15 that ∼50% of the NMOC emitted by BB are unidentified on a mass basis and this should perhaps be a standard assumption in the modeling of BB plumes. The rate of increase in ∆O 3 /∆CO seen in the Yucatan and Africa is faster than observed in some BB plumes; especially at high latitudes (Goode et al., 2000;De Gouw et al., 2006), but most BB occurs in the tropics. In any case, from southern Africa to Alaska it has 20 been shown that large-scale chemical changes can occur in BB plumes in an initial photochemical regime that is different from the ambient boundary layer and at a spatial scale that could challenge regional-global models.
In Fig. 4a, and some of the figures that follow, there is some non-monotonic structure and/or "scatter" in the downwind normalized excess mixing ratios (NEMR). This is 25 expected due to several factors including: (1) Different parts of the plume can age at different rates. An aircraft cannot fly exactly down the plume center-line and instead goes in and out of regions where the smoke is more concentrated and there is less light. Also, the combustion rate at the source can fluctuate leading to a non-monotonic  (2) The fuels and initial emissions can vary over the course of a fire. When a fire burns freely into homogeneous fuels, the flaming/smoldering ratio and the initial emission ratios may be fairly constant , but this is not always the case. (3) Mixing with fresh or aging plumes from other fires is possible. ∆HCN/∆CO is one of the ER that varies the most from fire to 5 fire and this NEMR was fairly constant for the 1.5 h of Fire #3 data we show. (Figure 1b shows all the HCN and CO values in the aging Fire #3 plume.) However, some degree of mixing with other plumes cannot be completely ruled out. A rigorous error estimate is not possible for each of the above terms or the assumption of a similar windspeed before our sampling. Thus we point out obvious trends in 10 the data and, in some cases, we fit a line to the data and compare the slope to the standard error in the slope to determine if there is a statistically significant trend. The fractional uncertainty in the rate of any process discussed is larger than the standard error divided by the slope due to the additional uncertainty in the sample ages. Probably all the samples have experienced more aging, or all the samples have experienced 15 less aging, than estimated. The real uncertainty in the rate is probably about a factor of two. Figure 4b suggests that there were likely some gradual changes in the initial emissions of Fire #3 and the post-emission processing environment that help interpret our data in the aging plume. Figure 4b shows ∆BC/∆CO at the 5 s time resolution of the 20 PSAP. BC is a flaming product and CO is a smoldering product. As the C-130 flew downwind in the plume, the gradual 20% decrease in ∆BC/∆CO until about 0.9 h suggests the instruments were sampling smoke originally produced at a gradually decreasing flaming/smoldering (F/S) ratio. The peak in BC/CO at ∼1 h could reflect a temporary increase in the F/S ratio at the source about an hour before the sampling started.

25
For the older samples, the F/S ratio was again close to the F/S ratio at the time of the first C-130 sample. Thus a comparison of the beginning and end NEMRs may best reflect post-emission chemistry. The variation in J NO 2 is also shown in Fig. 4b Interactive Discussion J NO 2 decreases by ∼2, ∆O 3 /∆CO decreases slightly, and CO increases. The aircraft is evidently entering a region of the plume with greater total smoke concentration. After one hour, both ∆O 3 /∆ CO and J NO 2 increase, the smoke concentration decreases, and minimal cloud-processing may occur.
A key driver for photochemistry besides UV is OH. In Hobbs et al. (2003) the rate of 5 decrease of numerous NMHC in one African biomass burning plume was used to estimate an average plume OH over the first 40 min of aging of ∼1.7×10 7 molecules/cm 3 .
On our Fire #3, no NMHC were measured within the aging plume, but an OH instrument was on board. The first OH value in the aging plume (averaged over 29 s of flight time and a calculated range of smoke aging of 22-43 min) is 1.14×10 7 molecules/cm 3 .

10
This is 5-20 times larger than the OH values in nearby background air. The plume OH levels thereafter decreased to about twice the average OH in the boundary layer. To our knowledge this is the first in-situ measurement of OH in a BB plume and it confirms the potential for very high initial OH in BB plumes. The measurements imply a major shortening of reactive species "lifetimes" in comparison to ambient air as predicted ear- 15 lier (Mason et al., 2001). HO 2 and RO 2 would likely be elevated along with OH (Mason et al., 2001). However, the one minute time resolution of the HO 2 instrument and some missing data make it difficult to determine the levels of this species in the 3 BB plumes sampled on the C-130. Trace gases other than O 3 also increased. The initial ∆HCHO/∆CO from the NCAR 20 DFG spectrometer was 0.025±0.01 and it increased to 0.038±0.01 at 0.8±0.1 h. Similar increases in this NEMR were also seen in an African plume in samples acquired near the top of the plume . The ∆HCHO/∆CO increased dramatically when smoke entered a cloud in Africa (Tabazadeh et al., 2004) and may have also increased strongly in the Fire 3 plume for the points that may be cloud impacted 25 (∼0.065±0.025). The ratio ∆H 2 O 2 /∆CO was 0.0013 in the nascent smoke from Fire #3 and then increased by ∼4 to 0.0054 after ∼1.3 h of aging (Fig. 5). In Fig. 5 Figure 6 shows post-emission growth in peroxyacyl nitrates both as ∆PAN/∆CO and ∆ΣPANs/∆NOy. An initial value for ∆PAN or ∆ΣPAN may not be meaningful (as for O 3 ) and was not measured due to interference in the nascent smoke. However, Fig. 6a shows that the ∆PAN/∆CO ratio increases rapidly from ∼0.0025 (at ∼0.4 h) to ∼0.006 10 (at ∼1.4 h). The NEMR reached in ∼1.4 h is as large as the NEMR observed in smoke from Canada that was ∼8 days old during NEAQS (F. Flocke private communication). This demonstrates the large variability in both initial emissions and photochemical rates that are associated with BB plumes. Figure 6b shows the ∆ΣPANs/∆NOy with aging in the Fire #3 plume. In the ∼1.2−1.4 h aging interval, ∆PAN/∆NO y alone has increased 15 to about 13%. The other PAN-like species showed similar trends, but were present in smaller amounts. The sum of the most abundant other PAN-like species (APAN and PPN) was about 20% of PAN in the 1.2-1.4 h interval and the ∆ΣPANs/∆NO y had increased to 0.167±0.036 in this interval. In the initial Fire #3 plume ∆NO x /∆NO y was 0.76 (based on comparing integrals) and in the 1. The post-emission fate of organic acids in BB plumes is variable. Goode et al. (2000) observed secondary increases in both excess formic acid (∆FAc) and excess acetic acid (∆HAc) ratioed to ∆CO for one Alaskan fire, but only in ∆FAc/∆CO for another Alaskan fire. In 3 different African BB plumes, Yokelson et al. (2003) observed ∆HAc/∆CO increase to 6-9% in <1 h, but saw no post-emission increase in 5 ∆FAc/∆CO. A huge regional plume emanating from the Brazilian Amazon had much higher ∆FAc/∆CO (∼1.6%) and ∆HAc/∆CO (∼5%) than the initial values measured for Amazonian fires (Yokelson et al., 2007b). Within the aged Fire #3 plume, we measured more organic acids than previously possible. The excess mixing ratios of FAc and peroxyacetic acid normalized to ∆CO both increased (Fig. 7) while the excess 10 mixing ratios of HAc and propanoic acid normalized to ∆CO did not show significant trends. In general, the ratio of excess carbon contained in organic acids to excess carbon contained in CO is large in fresh smoke and very large in slightly aged smoke (4-18% in the Yucatan, Brazil, and Africa). Thus, in many BB plumes organic acids are the third most important carbon reservoir behind CO 2 and CO and ahead of CH 4 . 15 The observed growth in organic acids is often much larger than the amount of known precursors. There may be currently unidentified high molecular weight precursors that are emitted in quite variable amounts.
The total NO x loss for the 1.2-1.4 h aging interval was 54±19% with the formation of PANs accounting for a NO x loss of ∼17% as shown earlier. The measured OH 20 (starting at 1.1×10 7 and dropping to 4.1×10 6 ) coupled with the pseudo-second order rate constant for OH+NO 2 (1.02×10 −11 cm 3 /molecule/s) at the temperature and pressure of the plume (288 K, 815 hPa) predicts an additional 30% loss (Sander et al., 2006). The additive loss from these two channels (47%) is close to the total loss considering the uncertainty in the sample ages (a factor of two) and the rate constant 25 (∼30%), but other loss processes are not excluded. For the 22 March Twin Otter Fire #8, ∆NO x /∆CO decreased ∼75% between the young and aged samples. This larger decrease could be partly due to changes in this ratio at the source, but may also reflect the fact that the Twin Otter sampled initial smoke at lower altitude (300-600 m) than the

Tables Figures
Back Close

Printer-friendly Version
Interactive Discussion C-130 (∼1700 m). OH may be higher in the fresher smoke . Figure 8 shows ∆SO 2 /∆CO from both the NOAA UV and the Caltech CIMS in the C-130 Fire #3 plume. The ratio drops from the initial average of 0.0239 to ∼0.015 by ∼0.7 h. A decrease this fast is not likely to be due only to gas-phase or heterogeneous processes in the aging plume. The fact that SO 2 is produced by flaming combustion 5 and CO is mainly from smoldering combustion probably explains much of the observed drop. As shown in Fig. 4b, the samples with nominal aging times up to almost 1 h reflected relatively more smoldering at the source. For the 1.2-1.4 h aging interval, the F/S was about the same as the initial value and ∆SO 2 /∆CO was 0.0186±0.006 on the continuous instrument (no CIMS samples). This could imply an SO 2 loss of 23±8% 10 due mainly to plume processes. The expected loss by OH oxidation is only ∼3% at 1.3±0.1 h. According to Seinfeld and Pandis (2006), SO 2 loss is very fast in cloud droplets and also significant on particles at RH>90% or near NH 3 sources, conditions which are met for much of the Fire #3 plume. The expected and observed changes in sulfur species over the relatively short time monitored in the Fire #3 plume are small and uncertain, but heterogeneous processes may well have contributed.
A large increase in ∆PM 2.5 /∆CO was observed in the aging Fire 3 plume. Figure 9a shows a rapid initial rise in ∆PM 2.5 /∆CO, which may partly reflect decreased F/S at the source combined with the higher EFPM normally observed for smoldering (Yokelson et al., 2007a). At ∼1.4 h of aging the ∆PM 2.5 /∆CO had increased by a factor of 20 2.6±1.5, which is independent of the choice of a fixed MSE. Figure 9b shows the mass fraction ∆BC/∆PM 2.5 , which decreased by 2.6 over the same aging period. This decrease is independent of the choice of fixed MAE. Since BC is made only by flaming combustion at the source and ∆BC/∆CO is similar at the source and the end of the aging period, the complementary change in the above ratios confirms that the overall 25 change in ∆PM 2.5 /∆CO is due to secondary aerosol formation. Also shown in Fig. 9b is the increase in SSA as the mass fraction of BC decreases in the PM 2.5 . Secondary formation of the individual non-absorbing constituents was quantified by the AMS as described next.  Figure 10 shows the Fire #3 aging results for all the particle species measured by the AMS except for NR chloride. The excess amounts of all the particle species shown are normalized to excess CO and all these ratios increased significantly. (∆chloride/∆CO remained constant over ∼1.4 h). Summing the ER to CO (on a mass basis) in the fresh smoke of Fire #3 gives a total of 5.6%. We derived a "final" mass ER to CO from the 5 fit shown at 1.4 h and then summed these ratios to obtain 13.3%. Thus, the excess mass of all AMS species ratioed to excess CO increases by a factor of 2.4 over this time period -in good agreement with the factor of 2.6 implied by the light scattering.
The changes in the mass percentage of the ∆PM 2.5 were also computed for each species yielding complementary results since the particles were adding mass. For in-10 stance, the ∆sulfate/∆CO ratio increased by a factor of 8.6±2.7 (95% CI) from its initial value to its value at 1.4 h. This was larger than the overall PM 2.5 growth factor of ∼2.6 so the mass percentage of the particles accounted for by sulfate increased from ∼1.5 to ∼5%. The growth factor measured for ∆OA/∆CO was 2.3±0.85, which is slightly smaller than the overall PM 2.5 growth of ∼2.6 so the calculated mass fraction of OA in 15 the PM 2.5 decreased slightly from 62 to 53%. Disregarding uncertainty, the calculation suggests that the particles could be adding organic mass, but adding inorganic mass at an even higher rate. The growth factor for ∆NO − 3 /∆CO was 5.4±1.2 and the mass percentage of nitrate increased from 5.5 to 11%. The ratio ∆NH + 4 /∆CO increased by 2.7±2.0 and so the computed mass percentage of ammonium was nearly unchanged 20 (5.3% initial to 5.4% final). Since ∆Cl − /∆CO (final) divided by ∆Cl − /∆CO (initial) was 1.02, the calculated mass percentage of Cl − decreased by 2.6 (from 14.7% to 5.7%); the inverse of the overall particle growth. Note in this case Cl − was not lost from the particles; it simply represented a decreasing fraction of the PM. The AMS spectra showed that the ∆OA/∆OC ratio increased with aging (Fig. 10 Interactive Discussion accounts for 49±16% of the decrease in ∆NO x /∆CO expected due to OH reaction. The molar increase in ∆sulfate/∆CO accounts for 39±16% of the total observed ∆SO 2 /∆CO decrease. Similar calculations for ammonium are not possible since NH 3 was not measured on the C-130. However, on a molar basis, the ammonium formation is nearly equal to the sum of the sulfate and nitrate formation. A balance between 5 the loss of gas phase organic precursors emitted by BB and the secondary formation of BBOA is complex for several reasons. The condensed phase species comprising BBOA are mostly unknown and they likely added mass during gas-phase oxidation. In addition, only ∼50% of the total gas-phase NMOC emitted by BB have been identified. Grieshop et al. (2008) recently found that the loss of known secondary OA (SOA) 10 precursors in wood stove smoke only accounted for ∼20% of the measured SOA formation in a smog chamber experiment. They also developed a model that accounted for all the observed SOA formation by assuming that unspecified low volatility organic compounds were present in significant amounts; an assumption that is not included in other current SOA models. Finally, in real BB plumes some of the secondary aerosol 15 formation can occur due to condensation of trace gases from other sources. However, we get a rough idea of the magnitude of the conversion process as follows. We estimate the growth for each AMS particle species by applying the measured growth factor from Fire #3 to the average initial EF for that species shown in Table 4. We then divide by the average initial EF in Table 4 for the trace gas precursor to get an upper limit 20 for the fractional precursor conversion. With these assumptions, the mass of the PM species formed in ∼1.4 h divided by the mass of the co-emitted gas-phase precursor is: sulfate/SO 2 (0.11), OA/ΣNMOC (0.19), NO − 3 /NO x (0.26), and NH + 4 /NH 3 (0.32). The OA/NMOC fraction is overestimated as explained above: i.e. the projected mass yield of OA at 1.4 h is 0.10 compared to estimated total NMOC (including unmeasured 25 species). However, the growth factors observed in this work, coupled with the TEM results presented next may be the most comprehensive evidence available for rapid secondary formation of both organic and inorganic aerosol in a BB plume. The large amount of secondary formation observed in this work has important implications for air 799 Introduction

Tables Figures
Back Close

Full Screen / Esc
Printer-friendly Version Interactive Discussion quality, radiative transfer, and visibility. This is especially true because biomass burning is the second largest global source of NMOC (∼500 Tg/yr) after biogenic emissions (∼1000 Tg/yr) (Yokelson et al., 2008). Finally, large formation of SOA may not always occur in BB plumes (Capes et al., 2008) so this subject deserves significant further research.

5
TEM analysis of individual aerosol particles shows that their composition and structure in the smoke changed within 30 min of emission, presumably through condensation, heterogeneous reactions, or both (Fig. 11). We compared TEM samples that were collected at two altitudes (∼600 m and ∼1700 m) on the Twin Otter from the 22 March Fire #8. The younger sample was collected within a few minutes of emission. The older 10 sample had aged between 10 and 30 min since emission.
The younger sample contains many particles with a large fraction of KCl, whereas the older one has particles that contain K, S, and Na and that presumably consist of K 2 SO 4 , Na 2 SO 4 , and their mixtures. The older sample also includes spherical organic aerosol particles (tar balls; indicated with a "T" in Fig. 11c) like those reported by Pósfai 15 et al. (2004) and (Hand et al., 2005).
The increases in S and organic C with aging measured by using the TEM are consistent with the increases for the same species measured using the AMS on the C-130 Fire #3. One possible difference between the two fires is that the TEM data show that there is almost no Cl in the 10-30 min old sample whereas measurements with 20 the AMS revealed a decrease in the mass percentage of chlorine, but a constant chlorine/CO ratio with aging. Chlorine could decrease with age as the result of reactions in the atmosphere. Similar Cl loss was also observed in African biomass burning plumes (Pósfai et al., 2003;Li et al., 2003). KCl reacts with sulfuric acid and forms K 2 SO 4 , releasing Cl as HCl. A possible reason for the difference in the results for the 25 two fires could be that Cl loss is a fast initial process and the youngest AMS sample was at the same altitude as the oldest TEM sample. If most of the KCl particles reacted within 10 to 30 min of emission, no further Cl decrease would have been detected with AMS in their biomass burning plume. Another possibility is that as the sulfate is replac- ing the chloride in KCl, the chloride is recondensing as NH 4 Cl, which is detectable with the AMS but is likely lost in the vacuum of the TEM (Johnson et al., 2008). In any case, to our knowledge, Cl − has been observed to decrease or be conserved as BB plumes age, but not to increase.
3.5 Chemistry of aged smoke layers at top of boundary layer 5 Unlike the case for C-130 Fire #3, we don't know the initial ER for the fires that contributed to the aged, and potentially cloud-processed, smoke layers that were penetrated at the top of the BL on 23 March. However, we can compare the normalized excess mixing ratios in these high-altitude, aged smoke layers to our study-average initial ER measured in nascent smoke and the trends observed in the aging of the 10 Fire #3 plume to gain some additional insight into the likely nature of smoke aging. The general observation is that these layers tend to have normalized excess mixing ratios similar to those in the aged Fire #3 plume or exhibit a continuation of the aging trends noted in the plume downwind of Fire #3. But some caveats are noted. The ∆O 3 /∆CO ratio in these layers increases with altitude from ∼0.02 to ∼0.05 15 to 0.17 ( In the smoke layers at the top of the BL ∆NO x /∆NO y ranged from 0.21 to 0.37, which are all values lower than the average (0.46) in the Fire #3 plume at ∼1.4 h. This suggests that the smoke had aged for more than ∼2 h. However, the lowest value (indicating the oldest layer) was in the middle layer rather than the top layer, which ∆O 3 /∆CO indicated was the oldest. ∆ΣPANs/∆NO y was the highest in the bottom 5 layer (∼0.3) and <0.15 in the other layers. This would imply that the bottom layer is the oldest. Thus, several normalized excess mixing ratios that can be used as "photochemical clocks" do not all indicate the same average age or relative ages -possibly due to varying initial emissions, mixing of different aged plumes, and different processing environments. An important point of this discussion is that given fire to fire variability (Fig. 3) and the high uncertainty in mean EF (Table 2) it could be misleading to compare aged and fresh smoke from different fires even when they occur or originate in the same region (Capes et al., 2008). Conversely, a non-representative characterization of the average initial BB emissions from a region, even if coupled with adequate knowledge of the post-emission transformations (which is unlikely), could cause erroneous 15 calculations of the speciation of the BB emissions exported from a region. We discuss the representativeness of our initial emissions measurements and the topic of scaling to regional emissions next.
4 Implications and impact on the regional environment 4.1 Fire characteristics and estimates of regional emissions 20 It is important to characterize the fires we sampled, consider how well they represented the regional biomass burning, and help guide reasonable estimates of regional emissions from our spot measurements. In this effort, we rely on notes and photographs from the sampling aircraft, the literature, and remote-sensing data. Based on visual observations from the Twin Otter, nearly all the fires fit into a simple classification scheme: (2) Burning of crop residue (CR) or "agricultural waste" in fields. (3) DF or CR fires which had "escaped" into nearby un-slashed forest where they consumed mostly understory fuels. If the unintended burned area was <∼10% of the planned burn, the fire was classified as DF or CR in Tables 1 and 2. If the planned and unplanned areas were comparable the fire was classified as "mixed." 5 Surface windspeeds were high throughout southern Mexico in March of 2006 and at least one-third of the fires we observed from the Twin Otter had escaped their apparent intended boundaries. The fact that much of the Yucatan forest canopy is leafless in the dry season also promotes large amounts of unintended burning; in contrast to evergreen forests where the intact canopy maintains a moist microclimate during the 10 shorter "dry"/fire season. During our flights a total of 6 DF, 6 CR, 5 mixed, and 3 fires of unknown type were observed. This suggested that the clearing of new areas and the disposal of crop waste were responsible for a roughly equal number of fires. In addition, we estimated the area burned and fuel consumption for the whole Yucatan region using a model Interactive Discussion derstory fuels were burning in tropical dry forests in Brazil and Africa (Ward et al., 1992;Shea et al., 1996;Hoffa et al., 1999). For crop waste fires total biomass and the percentage that constitutes residue varies considerably by crop. Further, the percentage of residue that decomposes before burning varies with climate and drying time and use of residue as fertilizer, animal feed, or household fuel also varies (Yevich and Logan, . We could not identify the plant material that was burned in this study. However, cereals are the most common crop worldwide and data from Zárate et al. (2004) suggests a range of fuel consumption of 0.36-4.2 Mg/ha for cereal crop residue burning. Higher values have been measured for other crops that are important in Mexico with the largest values typically being reported for sugar cane (20 Mg/ha;Lara et al., 2005). 10 Thus, the NCAR model averages lie well within the rather large range of values that occur for these fire types.
The uncertainty in the MODIS-based area burned has two chief components: detection efficiency and the estimation of burned area from counts of fire pixels. The MODIS hotspots show daily trends in fire frequency and the spatial distribution of biomass 15 burning, but not all fires are detected. The timing of the MODIS overpasses, closely spaced (∼90 min) midday and nighttime pairs, leads to omission errors for short-lived fires or fire activity initiated following the last daytime pass. There are also times when clouds preclude fire detection or overpasses do not cover an area of interest. Additionally, the detection rate decreases with burn size, Hawbaker et al. (2006) found the 20 MODIS detection rate crossed the 50% mark at a fire size of ∼100 ha. In fact, only a small fraction (4/20) of the fires we sampled were detected by MODIS ( Table 1). Most of the "missed fires" were sampled after the last MODIS overpass and all of the fires we photographed were <100 ha. Missing fires would cause the model to underestimate burned area. On the other hand, all the fires that we could roughly size were 25 significantly smaller than the ∼100 ha size assumed in the model. In addition, Giglio et al. (2006)  Interactive Discussion of error in area burned tend to offset. In light of the above-discussed challenges, the overall model uncertainty in the total amount of emissions is estimated at a factor of two (Wiedinmyer et al., 2006). The results of applying the NCAR model to the Yucatan peninsula (∼18−22 N by 87−91 W) for March through May of 2006 are shown in Table 6. The ratio of area 5 burned in forest to croplands is about 6:1 for March 2006 and 7.5:1 for the entire March-May 2006 biomass burning season. This contrasts with our airborne observation that the two fire types occurred in roughly equal numbers. Because of the higher fuel consumption per unit area for forest burning, forest fires account for nearly all the modeled fuel consumption on a mass basis. We compared the NCAR model result 10 for March 2006 with a separate analysis using a different vegetation map (CONABIO, 1999) and the ratio of MODIS active fire detections in forested areas to agricultural areas was in rough agreement at 4:1. The predominance of fire in forested areas in the MODIS based analyses compared to the airborne sampling may reflect several factors. 1) The airborne sampling was concentrated nearer the coast (Fig. 2) while the fraction Interactive Discussion for DF/CR may be an upper limit as it represents approximately doubling the area under agriculture annually. This upper limit would imply an approximately 11:1 ratio in fuel consumption, which suggests our simple average of all ER presented in Table 4 might ultimately be improved upon.
4.2 Regional transport and estimated total production of emissions 5 As discussed above, the fuels and the mix of flaming and smoldering impact the initial emissions from fires, which can then evolve rapidly. The post-emission transport dictates what areas of the atmosphere are impacted by the emissions and the photochemical processing environment. Figure 12 shows 5-day HYSPLIT forward trajectories from 3 starting altitudes from each fire we sampled (Draxler and Rolph, 2003). 10 Similar forward trajectories were confirmed for multiple days in April and May of 2006, which are the peak biomass burning months. About one-half of the emissions tend to disperse northwesterly and remain at altitudes that are likely within the boundary layer. These trajectories tend to cross under NE trending trajectories from MC over the Gulf of Mexico. There is potential for the Yucatan and MC plumes to eventually mix (after 15 some evolution of each plume) over the US. Forward trajectories that start in the early evening (not shown) can disperse directly to the west towards Mexico City. Much of that transport will happen at night. In any case, it's possible for Yucatan emissions to reach Mexico City, other areas in Mexico, or a large area of the Midwestern and eastern US as has been documented in detail elsewhere (Lyons et al., 1998;Kreidenweis 20 et al., 2001;Rogers and Bowman, 2001;Bravo et al., 2002;Lee et al., 2006;Massie et al., 2006). Because a large area is impacted by Yucatan BB emissions we next discuss the amount of these emissions as calculated by the NCAR model described above (Table 6). To help place the estimated amount of Yucatan emissions in perspective we  amounts of other trace gases as the MCMA. However, the PM 2.5 in the MCMAEI could be more than a factor of 2 low (calculated from Salcedo et al., 2006). In addition, the simple comparison in Table 6 does not include secondary aerosol formation, which could increase the amount of PM 2.5 tenfold in the MCMA emissions (Kleinman et al., 2008) and a factor of two or more in the BB emissions (Sect. 3.4). Adjusting for the 10 above factors implies that the Yucatan SO 2 and PM 2.5 sources are likely about 2 and 4 times larger, respectively, than MCMA on an annual basis. Additional important pollution sources are located in the region including fires in other parts of Mexico and Central America and large SO 2 emissions from the Popocatépetl volcano (894 000 t yr −1 , Grutter et al., 2008), the Tula refinery (158 000 t yr −1 , de Foy et al., 2007), and other vol- 15 canoes and refinery complexes. Notably, both the Popocatépetl and Tula SO 2 sources are estimated to be larger than the Yucatan BB or MCMA SO 2 source. Clearly though, our BB estimates for the Yucatan establish it as one of the major regional sources of trace gases and particles.  .

Conclusions
This study increased both the total number of fires sampled (from 2 to 22) and the total number of plume species quantified (from 4 to 51) for biomass burning (BB) in the Yucatan. These are the first detailed measurements of NH tropical BB and they include 5 some species measured for the first time in young plumes (OH, peroxyacetic acid, propanoic acid, hydrogen peroxide, etc Cofer III., W. R., Levine, J. S., Winstead, E. L., Stocks, B. J., Cahoon, D. R., and Pinto, J. P.: Trace gas emissions from tropical biomass fires: Yucatan peninsula, Mexico, Atmos. Environ., 27a, 1903Environ., 27a, -1907Environ., 27a, , 1993 Comisión Nacional para el Conocimiento y Uso de la Biodiversidad (CONABIO): Uso de suelo y vegetación modificado por CONABIO, Escala 1:1000000, Comisión Nacional para el Conocimiento y Uso de la Biodiversidad, Mexico City, Mexico, 1999. Crounse, J. D., McKinney, K. A., Kwan, A. J., and Wennberg, P. O.: Measurement of gas-phase hydroperoxides by chemical ionization mass spectrometry, Anal. Chem.,78(19), 6726-6732, 10 2006. Crutzen, P. J., Lawrence, M. G., and Pöschl, U.: On the background photochemistry of tropospheric ozone, Tellus B, 51, 123-146, 1999. DeCarlo, P. F., Kimmel, J. R., Trimborn, A., et al.: Field-deployable, high-resolution, time-of-flight aerosol mass spectrometer, Anal. Chem., 78, 8281-8289, 2006.   :10.1029/2004JD004521, 2004 dissociation-chemical ionization mass spectrometry (TD-CIMS) technique for the simulta-5 neous measurement of peroxyacyl nitrates and dinitrogen pentoxide, J. Geophys. Res., 109, doi:10.1029/2004JD004670, 2004 Estimating the area of stubble burning from the number of active fires detected by satellite, Remote Sens. Environ., 109 (1), 95-106, 2007. Susott, R. A., Olbu, G. J., Baker, S. P., Ward, D. E., Kauffman, J. B., and Shea, R.: Carbon, hydrogen, nitrogen, and thermogravimetric analysis of tropical ecosystem biomass, in: Biomass Burning and Global Change, edited by: Levine, J. S., MIT Press, Cambridge, 350-360, 1996. Tabazadeh, A., Yokelson, R. J., Singh, H. B., Hobbs, P. V., Crawford, J. H., and Iraci, L.        The white arrows plus chemical symbols indicate particles that are rich in S, Si+S, or Al. T indicates tar ball particles. (d) EDS spectra of typical particles in image (c). We used an electron beam ∼0.2 µm in diameter and 30-s counting times for the EDS analyses. The spectra were averaged across the particles indicated by arrows in the images. Almost all particles included strong C and O signals from organic material and the lacey-carbon filter. N-bearing species such as nitrates and ammonium were not well measured using EDS because N signals are generally weak plus these species decompose rapidly in the TEM beam. The Cu peak is marked with an asterisk because it is from the TEM grid. Sample collection times were less than a minute for both samples. Scale bars in both images indicate 1 µm. Org indicates organic material. 9,2009 Emissions from biomass burning in the Yucatan   Table 1. The starting altitudes above ground level are blue (250 m), green (1000 m), and red (2500 m). The projected altitudes stay mainly within the boundary layer. The hotspots (red dots, no border) due to biomass burning throughout Mexico are shown for the dates sampling was conducted in this work.