Trace gas and particle emissions from domestic and industrial biofuel use and garbage burning in central Mexico

. In central Mexico during the spring of 2007 we measured the initial emissions of 12 gases and the aerosol speciation for elemental and organic carbon (EC, OC), anhydrosugars, Cl − , NO − 3 , and 20 metals from 10 cooking ﬁres, four garbage ﬁres, three brick making kilns, three charcoal making kilns, and two crop residue ﬁres. Global biofuel use has been estimated at over 2600 Tg/y. With several simple case studies we show that cooking ﬁres can be a major, or the major, source of several gases and ﬁne particles in developing countries. Insulated cook stoves with chimneys were earlier shown to reduce indoor air pollution and the fuel use per cooking task. We conﬁrm that they also reduce the emissions of VOC pollutants per mass of fuel burned by about half. We did not detect HCN emissions from cooking ﬁres in Mexico or Africa. Thus, if regional


Introduction
In developed countries most of the urban combustion emissions are due to burning fossil fuels. Fossil fuel emissions are also a major fraction of the air pollution in the urban areas of developing countries. However, in the developing world, the urban 15 regions also have embedded within them numerous, small-scale, loosely regulated combustion sources due to domestic and industrial use of biomass fuel (biofuel) and the burning of garbage and crop residues. The detailed chemistry of the emissions from these sources has not been available and the degree to which these emissions affect air chemistry in urban regions of the developing world has been difficult to as-20 sess. As an example, we note that Raga et al. (2001) reviewed 40 years of air quality measurements in Mexico City (MC) and concluded that more work was needed on source characterization of non fossil-fuel combustion sources before more effective air pollution mitigation strategies could be implemented. The 2003 MCMA (Mexico City Metropolitan Area) campaign (Molina et al., 2007)

and the 2006 MILAGRO (Mega-
The burning of crop residue in fields is generally considered to be the fourth largest type of global biomass burning with estimates including 540 Tg/y (Andreae and Merlet, 2001) and 475 Tg/y (Bond et al., 2004). Because cities are often located in prime agricultural regions, they may expand into areas where crop residue burning is a major activity and is sometimes the dominant local source of air pollution (Cançado et al., Introduction Conclusions References Tables  Figures   Back  Close Full Screen / Esc

Printer-friendly Version
Interactive Discussion in Table 1. These include eight indoor open wood cooking fires, two indoor wood cooking fires in Patsari stoves, three charcoal making kilns (from two sites), three brick making kilns, four garbage burns in peri-urban landfills, and two barley stubble field burns. All but one of the open wood cooking fire measurements were conducted in rural and 5 semi-rural homes during actual cooking episodes. The cooking fire in the laboratory of the Interdisciplinary Group on Appropriate Rural Technology (GIRA) was a simulation using an authentic open cook stove and typical fuel wood. For six of the eight homes in which we sampled, the kitchen was housed in a separate building. For the other two, the kitchen was part of the main dwelling with a wall separating it from the sleeping 10 area. Ventilation in all cases was by passive draft through door and window openings, cracks in the walls between boards, and horizontal openings where roof meets wall. Six of the eight kitchens had a dirt floor, seven were constructed of wood and one of brick. A variety of biofuels were available to the homeowners, including wood, corn cobs, corn stalks, and charcoal. The primary fuel in all these homes, and the fuel 15 used in all the fires we measured, was oak or pine collected locally by hand. Cooking fires were built either directly on the ground within a ring of three rocks, or on a mud and mortar, u-shaped, raised open stove. In one instance the "stove" was a dirt-filled metal bucket with rocks on top. A typical food preparation regimen begins with a small, hot, flaming fire to quickly boil a pot of water, which is then loaded with beans and 20 set off to the side to simmer. As the fire begins to die back, the cook begins frying tortillas. Wood is fed gradually to the fire to maintain the right amount of heat and when the cooking ends the fire is generally snuffed out to conserve fuel. A cooking session might last several hours depending on how much food is needed in the next few days. The sample lines of all the instruments were co-located at ∼1 m above the fire over the 25 course of the cooking operation. The cook and her youngest children typically remain inside the kitchen for as long as it takes to prepare the food. The Patsari stove incorporates an insulated fire box that is vented to the outdoors by a metal chimney. It is the product of 15 years of work by GIRA  Ecosystems Research (CIECO) to improve stoves economically (Masera et al., 2005). The stove cuts fuel consumption "per cooking task" roughly in half so its widespread adoption could reduce the total emissions from biofuel use. The chimney provides an approximate 70% reduction in indoor air pollution (Zuk et al., 2007), which is the largest single factor causing mortality in children under five globally (Dherani et al., 5 2008). It is also of interest that reactions on the chimney surface could modify the emissions (Christian et al., 2007). The chimney does not eliminate all the indoor pollutants because the fire box has an open front that can leak emissions into the room. Also, the chimney emissions may at times be recirculated into the kitchen from outdoors. We sequentially measured first the kitchen air above the stove, and then the 10 chimney emissions from two different Patsari stoves in Pátzcuaro. One was located in a rural kitchen and the other was a newer model located in the GIRA lab's simulated kitchen.
We sampled three charcoal making kilns in a forested area between MC and Puebla. An excavation ∼5 m in diameter is dug by hand and kindling (dry needles, leaves, and 15 twigs) is laid down. Oak logs are stacked in the center and a network of interlaced green oak branches is placed over the top. The excavated dirt is then packed on top to complete the earthen kiln, which has about a dozen vents around the circumference. A kiln of this design yields 200-250 kg of charcoal in about eight days. The supporting oak branches burn away slowly and the kiln must be rebuilt once or more during its 20 lifetime to prevent it from collapsing and smothering the fire. The two kilns at the Hueyitlapichco site were constructed on consecutive days. We sampled them on their second and third day of operation on 19 April, and on their third and fourth day of operation on 20 April. At the San Gaspar site we sampled a single kiln on its fifth day of operation. Introduction

Conclusions
References Tables  Figures   Back  Close Full Screen / Esc

Printer-friendly Version
Interactive Discussion on the crossbeams (spaced to allow even heat circulation). Brick walls and a roof are then built up around the whole assembly. A fire is lit and fuel is shoveled or thrown in until the desired temperature is reached. Fuel is then added, as needed, to maintain that temperature around the clock for 2-3 days. At varying times each kiln operator uses mortar to seal the walls and most of the roof. Some owners allow the kiln to 5 ventilate freely through the walls and roof for a day before sealing with mortar, claiming this gives a more uniform bake. Others seal the walls and roof before ignition. Kilns number 1 and 2 were burning fuel that was mostly wood waste products that had been hauled onto the site by dump truck. About 90% of this fuel was sawdust by volume. The remainder was divided fairly evenly between wood scraps, plywood, and particle 10 board. A small fraction (less than 1%) was paper and cardboard. Brick kiln number 3 was using only scrap lumber while we made measurements. (We were unable to visit a fourth kiln near Silao that was reportedly burning used motor oil for fuel and a fifth kiln near Salamanca that was burning domestic waste scavenged from a nearby landfill.) The raw material for bricks is soil carved by hand from the ground in the vicinity of the 15 kiln. The soil is mixed with water and manure or other organic waste and stomped barefoot to form a thick paste. The paste is then pressed into a mold and overturned one by one into rows to dry in the sun. Once they are dry enough to handle, the green bricks are stacked (in the shade if possible) and covered to prevent too rapid drying and cracking. Two of the brick kilns were sized to fire 10-12 000 bricks at a time; the 20 third (brick kiln 2) was about three times larger. Kilns of this design are typical for Latin America and Africa, while more efficient designs -and coal fuel -are more common in Asia. All four garbage burning fires were in the municipal landfills of peri-urban communities north of Mexico City. Only one landfill (Coyotepec, garbage fire 2) was burning 25 when we arrived. This site was evidently excepted from a "no burn" policy that had been in place in northern Mexico state for some months. At the other three sites we ignited relatively small, representative sections of refuse under the direction of local authorities. The landfills held typical household and light industrial refuse. Plastic was by far the most abundant material present. The following list is an approximate accounting of the composition of the waste stream for these landfills, in roughly diminishing order: It appeared that tires were piled separately and perhaps not burned intentionally at the landfills. Wood was absent from any of the landfills since it is the most common cooking fuel in Mexico. The Tolcayuca landfill (garbage fire 3) was located in a textile 15 manufacturing area and contained a higher proportion of textile waste than the other landfills. Each landfill was attended by people who manually removed items of value, including recyclable plastic bottles and cardboard. The scavenging process was less than perfectly efficient and small portions of the "collectible" waste did get included in the burns. All of our measurements were made from fires burning in the processed 20 refuse from which the bulk of the recyclables had already been removed.
The agricultural waste burns took place in two adjacent, ∼2 ha barley fields northwest of Salamanca. The fields had been mechanically harvested so all that remained were 10109 Introduction Tables  Figures   Back  Close Full Screen / Esc

Printer-friendly Version
Interactive Discussion standing stalks (stubble, ∼15 cm) and a mat of broken stalks and chafe, all of it tinder dry. Photographs of many of the field sites described above can be found at http: //www.cas.umt.edu/chemistry/faculty/yokelson/galleries/album Mex/index.html.

Instrumentation
The primary instrument for measuring trace gas emissions was our mobile, rolling 5 cart-based Fourier transform infrared spectrometer (Fig. 2, Christian et al., 2007). It is rugged, easily transported, optionally self-powered, and can be wheeled to remote sampling sites. The optical bench is isolated from the chassis with wire rope shock absorbers (Aeroflex) and holds a MIDAC 2500 spectrometer, White cell (Infrared Analysis, path length 9 m), MCT detector (Graseby), and transfer and focusing optics 10 (Janos Technology tra are collected. We used nonlinear least squares, synthetic calibration (Griffith, 2002) to retrieve excess mixing ratios from the spectra for water (H 2 O), carbon dioxide (CO 2 ), carbon monoxide (CO), methanol (CH 3 OH), methane (CH 4 ), ethylene (C 2 H 4 ), propylene (C 3 H 6 ), acetylene (C 2 H 2 ), formaldehyde (HCHO), and hydrogen chloride (HCl). We used spectral subtraction (Yokelson et al., 1997) to retrieve excess mixing ratios for 20 CH 3 OH, C 2 H 4 , C 3 H 6 , C 2 H 2 , ammonia (NH 3 ), formic acid (HCOOH, also denoted FAc), and acetic acid (CH 3 COOH, also HAc). At a path length of 9 m the detection limit for most gases was ∼50-200 ppb. These gases accounted for all the quantifiable features in the IR spectra. The typical uncertainty for mixing ratios was ±10% (1σ). For CO 2 , CO, and CH 4 , the uncertainties were 3-5%. More complete descriptions of the system 25 and spectral analyses are given in Christian et al. (2007). After the campaign we checked for analyte losses in the FTIR cell due to adsorption or other reasons . The average NH 3 concentration in the cell 10110 Introduction Tables  Figures   Back  Close Full Screen / Esc

Printer-friendly Version
Interactive Discussion during one minute of signal averaging (the typical sampling time used in Mexico) was about 71% of its initial level. The average HCl was ∼93% of its initial level for the same interval. The ammonia and HCl results reported here have been adjusted upward to account for these cell losses.
A basic filter-sampling system with an internal pump (3 L min −1 ) and an impactor 5 was used to collect fire-integrated PM 2.5 on quartz filters. Analyses of the quartz filters were performed in the laboratories of the Research Center for Environmental Changes, Academia Sinica, Taipei, Taiwan. Organic and elemental carbon (OC, EC) were determined with a Sunset Laboratory, Inc. continuous carbon analyzer using thermal-optical transmission (Birch and Cary, 1996;Engling et al., 2006). Anhydrosug-10 ars (levoglucosan, mannosan, galactosan) were determined using high-performance anion-exchange chromatography (HPAEC) with pulsed amperometric detection (Engling et al., 2006). Soluble ions were determined with ion chromatography (Hsu et al., 2008b). We analyzed the quartz filters for trace elements using inductively coupled plasma-mass spectrometry (Hsu et al., 2008a;Hsu et al., 2008b). 15 We did not sample particles with Teflon filters, which are used for gravimetric determination of total PM 2.5 . However, we did deploy an integrating nephelometer (Radiance Research M903) that measured particle light-scattering at 530 nm and 1 Hz. The nephelometer was calibrated with particle free zero air and CO 2 before and after the campaign. The M903 nephelometer response was attenuated at the highest con-20 centrations we encountered in Mexico. Thus, we applied a correction factor to those high values based on direct comparison in laboratory smoke between the M903 and a TSI 3563 nephelometer, which does have a sufficiently large linear range. The M903 nephelometer output (b scat , m −1 ) has been compared directly to gravimetric PM 2.5 determinations on cooking fires in both Honduras (Roden et al., 2006) and Mexico (Brauer 25 et al., 1996). For dry, fine particles the conversion factor depends mostly on the EC/OC ratio of the particles. Our average EC/OC ratio (0.284) for cooking fires was very close to that reported by Roden et al. (2006)  Interactive Discussion light-scattering data from our cooking fires to an estimated total PM 2.5 as follows: b scat (530 nm, 273 K, 1 atm) × 552000 ± 75 000 = PM 2.5 (µg/m 3 , 273 K, 1 atm) (1) (The conversion factor is equivalent to a mass scattering efficiency of 1.8). This approach probably gives an uncertainty in our average PM 2.5 for cooking fires of about 20-30%. 5 The particles from the other fires were different enough that we did not estimate a total PM 2.5 for these sources based on light scattering. However, we do report the mass sum of the particle constituents on the quartz filters. In this sum, we multiply the OC by a conservative factor of 1.4 to account for non-carbon organic mass (Aiken et al., 2008). The species measured include most of the major particulate components with the exception of sulfate and ammonium, which accounted for only a few percent of particle mass in other Mexican biomass burning particles (Yokelson et al., 2009). Thus, the sum of detected species is likely not more than 10-30% lower than the total PM 2.5 .
We also deployed a CO 2 instrument (LICOR LI-7000) that was calibrated both be- 15 fore and after the campaign (negligible drift) with NIST-traceable standards spanning the CO 2 range encountered in the field. The CO 2 , nephelometer, and filter sampling systems shared a single inlet (conductive silicon tubing) that was often co-located with the FTIR sample line. In the cases where the FTIR mobility allowed sampling of the emissions at more points than the other instruments, the accurate determination of 20 CO 2 by both the LICOR and the FTIR allowed coupling the two data sets. CO 2 was also used to correlate the particle measurements to the trace gases measured by FTIR as described in detail elsewhere (Yokelson et al., 2009(Yokelson et al., , 2007.

Calculation of emission ratios and emission factors
An emission ratio (ER) is defined as the initial molar excess mixing ratio (EMR) of one Introduction Greek capitol delta -e.g. ∆CO, ∆CH 4 , ∆X, etc. Modified combustion efficiency (MCE) is defined as the ratio ∆CO 2 /(∆CO 2 +∆CO) and is useful for estimating the relative amounts of flaming and smoldering combustion during a fire, with high MCE corresponding to more flaming (Ward and Radke, 1993). To estimate the fire-average ER for a species "X" we plot ∆X for all the samples of the fire versus the simultaneously 5 measured ∆CO (or ∆CO 2 ) and fit a least squares line with the intercept forced to zero. The slope is taken as the best estimate of the ER as explained in more detail in Yokelson et al. (1999). Figure 1 is an example of this type of plot showing the CH 3 OH/CO ER derived from 10 FTIR samples obtained over the course of a wood cooking fire. An emission factor for any species "X" (EFX) is the mass of a species X emitted per unit mass of dry fuel burned (g compound per kg dry fuel). EF can be derived from a set of molar ER to CO 2 using the carbon mass balance method, which assumes that all of the burned carbon is volatilized and that all of the major carbon-containing species have been measured. It is also necessary to measure or estimate the carbon content of the fuel. For the fires using biomass fuel we assumed a dry, ash-free 15 carbon content of 50% by mass (Susott et al., 1996). For the garbage fires, which contained only some biomass, we estimated the relative abundance of the materials present from photographs. We then calculated the overall carbon fraction based on those proportions and carbon content estimates for each type of material (IPCC, 2006;USEPA, 2007). Table 2 shows that this procedure resulted in an overall carbon fraction 20 of 40% for the combustible landfill materials. The EF calculations for a charcoal kiln are complex because the fuel carbon fraction increases with time. We used a procedure identical to that described in detail by Bertschi et al. (2003). EFPM 2.5 for the cooking fires were calculated by multiplying the fire-integrated PM 2.5 to CO 2 mass ratio (gPM 2.5 /gCO 2 as measured by the nephelometer and LICOR) by the Introduction

Cooking fires
Trace gas ER and EF and particle EF based on light scattering for our cooking fires are given in Table 3. The first 10 columns of data are the eight open wood cooking fires plus a column each for the average and standard deviation. The next three columns 5 are the EF and average for the two Patsari stoves as sampled in the kitchen. The last three columns are the analogous data from the outdoor chimney exhaust of the same two Patsari stoves. The EF for individual particle species measured on the quartz filters are given for all the fires in Table 4. Open wood cooking fires are the main global type of biofuel use and we get an idea of the global variability in this source by comparing 10 EF from selected studies for some of the more commonly measured emissions (CO 2 , CO, CH 4 , and PM). Figure 2 shows EFCH 4 versus MCE (a function of CO and CO 2 ) for those studies, including this one, where CO, CO 2 , and CH 4 data were all available. (The data in Interactive Discussion the Ivory Coast. The Andreae and Merlet (2001) data point is a widely-used global estimate derived from the literature. The Bertschi et al. (2003) EFCH 4 appears higher than the trend and the Brocard et al. (1996) EFCH 4 lower, but these data are consistent with a tendency toward greater variability as the relative amount of smoldering emissions increases in biomass burning fires (Christian et al., 2007;Yokelson et al., 2008). 5 Particle EF also vary substantially with MCE as seen in Fig. 3, which includes EFPM from three of the same studies that are included in Fig. 2 (Andreae and Merlet, 2001;Johnson et al., 2008;Zhang et al., 2000), as well as two other relevant studies (Roden et al., 2006(Roden et al., , 2009). Roden et al. used a combination of nephelometry, absorption photometry, filter collection, and CO/CO 2 instrumentation to measure real-time and 10 fire-integrated EF from 56 fires in various stove types in rural Honduran homes, and 14 laboratory simulations in several stove types. Figure 3 incorporates only their data from 10 traditional, open wood cooking fires in homes. (CO 2 data for calculating MCE for the two Roden et al. (2006Roden et al. ( , 2009 studies were kindly provided by the authors.) Again there is considerable variability in EF for individual fires, but reasonable agree-15 ment between authors on the range and trend with MCE. This body of work on PM suggests a slightly lower average MCE (0.927) than implied in Fig. 2 As mentioned above, the use of improved stoves with chimneys and insulated fire boxes reduces both the total biofuel emissions (due to reduced fuel consumption) and the indoor air pollution. There is also potential for improved stoves to consume the fuel at higher MCE, reducing the EF for smoldering compounds. A further possibility is that 5 the surface of the chimney could scavenge some of the more reactive smoke components before they are emitted to the airshed. To examine these issues we compare the average MCE and EF of the Patsari chimney exhaust to the average MCE and EF for the open fire emissions. The average MCE was lower from our open fires (∼0.95) than it was from our Patsari chimney exhaust (0.97). Consistent with the increased 10 Patsari MCE, the EF for CO, CH 4 , and the measured NMOC (with the exception of organic acids, C 3 H 6 , and C 2 H 2 ) were about a factor of two lower from the chimney exhaust. For organic acids, NH 3 , and C 3 H 6 there was a larger drop (80-95%) in the EF measured from the chimneys that was likely due in large part to losses on the chimney walls. EFC 2 H 2 is similar for both sources as it is emitted by both flaming and 15 smoldering (Yokelson et al., 2008) and is not particularly "sticky." Overall, while only a fraction of the total NMOC emitted could be measured (Yokelson et al., 2008), the sum of the EFNMOC that were measured in this study from the chimney was ∼38% of the analogous sum from the open fires. We were unable to measure particle EF from the Patsari chimney. Johnson et al. (2008) (Patsari). They also reported a large reduction in the EF for CO, CH 4 , and PM, which was variable depending on the type of Patsari stove sampled. Based on the above, it appears that improved stoves could reduce both fuel consumption (by about half, Masera et al., 2005) and the amount of many pollutants emitted per unit mass of fuel consumed (by at least half).
There is a significant absence in Table 3 of HCN, which is widely used as a biomass burning tracer (Yokelson et al., 2007). HCN is normally well above the detection limits of our FTIR systems for landscape-scale biomass burning (e.g. forest fires, grass fires, ACPD 9, 2009 Trace gas and particle emissions in central Mexico  Yokelson et al., 2007). However, HCN was below our FTIR detection limits for cooking fires in both Africa  and Mexico (current study). A single FTIR sample from a Brazilian stove (Christian et al., 2007) did contain some HCN, but the ER to CO (0.0005) was ∼24 times lower than the value for Mexico City area forest fires (0.012, Yokelson et al., 2007). The low HCN/CO ER for cooking fires means that 5 where these fires are common, the biomass burning contribution to total pollution will be underestimated if it is based on an HCN/CO ER appropriate for landscape-scale burning (Yokelson et al., 2007). Acetonitrile is another useful biomass burning tracer (de Gouw et al., 2001), but cooking fire measurements for this species have not been attempted yet. However, 10 since acetonitrile emissions from other types of biomass burning are usually less than half the HCN emissions (Yokelson et al., 2009), they may also be unusually small from cooking fires. Methyl chloride (CH 3 Cl) has also been linked to biomass burning (Lobert et al., 1991), but its emissions are probably much smaller from cooking fires than for other types of biomass burning since wood has much lower chlorine content than other 15 components of vegetation (Table 4, Lobert et al., 1999). Levoglucosan and K (in fine particles) are also used as biomass burning indicators and they were observed in "normal" amounts in the particles from our cooking fires (Table 4) compared to other types of biomass burning. However, as discussed in more detail in Sect. 3.2, levoglucosan and K were also present in similar amounts in the fine particles from garbage burn-20 ing. Thus, in areas such as central Mexico where garbage burning is common it could contribute a significant fraction of the aerosol levoglucosan or K. The lack of a straightforward chemical tracer for cooking fires is especially significant since these fires will also not be detected from space as hotspots or burned area. In addition, the CO could be underestimated by MOPITT due to the low injection altitude for cooking fire 25 smoke (Emmons et al., 2004) and the short (one-month) lifetime for CO in the tropics. Thus, biomass burning estimates based on HCN or acetonitrile likely underestimate cooking fires (and total biomass burning), while estimates based on levoglucosan or K could be subject to "interference" from garbage burning in parts of the developing ACPD 9, 2009 Trace gas and particle emissions in central Mexico world. In summary, while survey-based research clearly indicates that biofuel use is the second-largest global type of biomass burning, there is not a simple chemical tracer to confirm this or to independently determine the amount of biofuel use embedded in urban areas of the developing world. 5 Our ER and EF for trace gases emitted by garbage burning are shown for individual fires in the left half of Table 5. Garbage fire 2 had already progressed to mostly smoldering combustion when we arrived. At the other three fires we sampled mostly flaming. Since we don't know the real overall ratio of flaming to smoldering combustion for landfill fires we just calculated the straight average and the standard deviation for 10 all four fires. For the trace gas EF this implies a 3:1 ratio for fuel consumption by flaming/smoldering, which does not seem unreasonable. The EF are computed assuming the waste in these landfills was 40% C by mass. If the %C is higher or lower the real EF would be higher or lower in direct proportion. It is important to note, however, that the ER to CO or CO 2 are independent of any assumptions about the composition of 15 the fuel. The EF for particle species are included in Table 4. Since we only have filter data for three of the four garbage fires, an average of the filter results is equivalent to one-third of the fuel being consumed by smoldering. We could not find any published, peer-reviewed, direct emissions measurements from open burning in landfills to compare our results to. Data from airborne and ground-20 based measurements of aerosols over the east Asian Pacific as part of ACE-Asia (Simoneit et al., 2004a, b) revealed significant levels of phthalates and n-alkanes in the aerosols. The presence of these compounds was attributed to refuse burning. A follow up study confirmed these compounds as major organic constituents in both solvent extracts of common plastics and the aerosols generated by burning the same plastics 25 in the laboratory (Simoneit et al., 2005). This indicated their potential usefulness as tracers. However, these are high molecular weight, semi-or non-volatile compounds whose relationship to volatile gaseous emissions is not known.

Garbage burning
10118 The comparison of the garbage burning emissions to biomass burning emissions is interesting. The average ethylene molar ER to CO for garbage burning (∆C 2 H 4 /∆CO, 0.044) is 3-4 times higher than for our open wood cooking fires (0.013, Table 3) or forest fires near Mexico City (0.011, Yokelson et al., 2007) and is likely a result of burning a high proportion of ethylene-based plastic polymer fuels. 5 HCl is not commonly detected from biomass burning (Lobert et al., 1999), but the EFHCl in the garbage burning emissions ranged from 1.65 to 9.8 g/kg, a range similar to that for CH 4 in biomass burning emissions. Lemieux et al. (2000) reported a strong dependence on PVC content for HCl emissions from simulations of domestic waste burning in barrels. Their EFHCl was 2.40 g/kg (n=2) for waste containing 4.5% PVC 10 by mass, and 0.28 g/kg (n=2) for waste with only 0.2% PVC. There was no mention of precautions taken to avoid passivation losses on sample lines, etc. (e.g. Yokelson et al., 2003). In the current study, significant additional chlorine was present in the particles; EF for soluble Cl − alone ranged from ∼0.2 to 1.03 g/kg fuel (Table 4). Studies of landfills in the European Union found that the chlorine content of solid waste was 15 about 9 g/kg (Mersiowsky et al., 1999) and that essentially all the chlorine was present as polyvinyl chloride (Costner, 2005), which is 57% Cl by mass. We found that burning "pure" PVC in our laboratory produced HCl/CO in molar ratios ranging from 5:1 to 10:1. Thus, the observed molar ER for HCl/CO in the MCMA landfill fires (0.037-0.19) are consistent with the burning materials we sampled containing ∼0.4-4% PVC. Our 20 results also suggest that the majority of the chlorine in burning PVC is emitted as HCl.
Even though the average EC/OC ratio for garbage burning (0.232, n=3) is close to that for the cooking fires (0.284, n=5), application of the cooking fire conversion factor to the garbage burning light scattering data underestimates the particle mass compared to summing the particle species data. Preliminary work in our lab suggests 25 this could be due to a shift to larger particles in the emissions from burning plastics. We can roughly estimate the EFPM 2.5 for garbage burning from the particle species data. The sum of the measured particle components averaged 8.74±7.35 g/kg, which, after allowing for unmeasured species, suggests that the EFPM 2.5 is about 10±5 g/kg.

ACPD
However, the levoglucosan emissions per unit mass of paper burned can be considerably higher than those from burning some types of biomass (Table 1, Simoneit et al., 1999). In our data, the average levoglucosan EF from garbage burning is 85% of the EF for cooking fires, which would make it difficult to use levoglucosan to distinguish between these two sources. The other sugars analyzed in this work (mannosan and 15 galactosan) showed more potential promise in this respect as their EF were ∼90% lower for garbage burning than for cooking fires. Finally, the garbage burning EF for mannosan was only ∼12% lower than the single mannosan EF measurement for crop residue burning. This tentatively leaves galactosan as the most promising sugar of those we analyzed to indicate general biomass burning in the presence of garbage 20 burning.
The garbage burning EF were the most different from the biomass burning EF for numerous metals. With correction for local soil composition, some of these metals could ultimately offer a useful method of assessing the garbage burning contribution to overall air quality. For example, the ratio EF garbage /EF cook for selected particle species 25 was: Sb (555.7), Pb (211.7), Sn (181.9), Cl − (63.7), Cd (33.57), As (20.9), Ca (5.1), and Mg (4.6). We note, however, that the soluble chloride in the one sample of crop residue burning smoke was actually higher than the average value for garbage burning. This could reflect the use of chlorine-containing agricultural chemicals (Sect. 3.4). In

ACPD Introduction
Conclusions References Tables  Figures   Back  Close Full Screen / Esc

Printer-friendly Version
Interactive Discussion examining the ratio of the average EF for garbage burning to the average EF for crop residue burning the most elevated metals are antimony and tin (Sb 309.4, Sn 33.6). Thus, initially Sb emerges as a promising tracer for garbage burning. Both Sb and PM 2.5 were measured in the MCMA ambient air at T 0 and T 1 during MILAGRO (Querol et al., 2008). The mean mass ratio for Sb/PM 2.5 for the March 2006 5 campaign at these sites was 0.000315. Our mean EF for Sb in PM 2.5 from pure garbage burning smoke was 0.011±0.008 g/kg. Our estimate of the average EFPM 2.5 for garbage burning is 10±5 g/kg, implying an Sb/PM 2.5 mean mass ratio of ∼0.0011 for this source. Comparison of the mean mass ratios of Sb/PM 2.5 for pure garbage burning and ambient air implies that garbage burning could account for about 29% of the PM 2.5 in the MCMA. However, we note that this estimate has high uncertainty and that the presence of Sb in the MCMA particulate was attributed by Querol et al. (2008) to the abrasion of tires and brake pads. In addition, higher Sb in the PM 2.5 might result if there were any manufacturing operations using Sb near the measurement sites. However, our initial crude estimate suggests that garbage burning deserves more attention as a 15 potentially significant contributor to the particle burden of the MCMA airshed. A more rigorous source attribution for garbage burning based on fine particle metal content would require a more complex multi-element approach. The main uses of antimony are as a flame retardant for textiles and in lead alloys used in batteries. Antimony trioxide is a catalyst that is often used in the production of polyethylene terephtha-20 late (PET) and that remains in the material. PET is the main material in soft drink bottles, polyester fiber for textiles, Dacron, and Mylar. The smoke particles from the dump with the highest percentage of textiles (Table 4, garbage fire 3) did have the highest mass percentage of Sb. We noted earlier that at least some of the PET materials (soft drink bottles) were being recycled rather than burned.

Brick making kilns
The particle and trace gas emissions data for brick kilns are in Tables 4 and 5, respectively. The brick kilns we sampled burned mostly biomass fuels and the identities of the emitted NMOC were similar to those from biomass burning. The brick kiln EF were 5 much reduced, likely due to the high MCE and to scavenging by the kiln walls and/or the bricks themselves. It is hard to say how well the emissions from these kilns represent brick making kilns in general because informal industries like brick kilns often burn a combination of biofuel, garbage, painted boards, tires, used motor oil, etc. Though our kilns burned mostly biofuel they emitted a much blacker smoke than any other 10 biomass burning we have observed (EC/OC 6.72, n=2). All the photographs of brick making kilns we took and could locate elsewhere showed very black smoke emissions. The high EFCl − , but low Sb and other metals for brick kiln 1 suggests that crop waste may have been a fuel component during our measurements or during past uses of the kiln. The elevated Pb from both kilns 1 and 2 may be due to burning painted boards 15 from demolished buildings. Painted boards were identified as a controversial fuel used in some Mexican brick kilns in a report to the USEPA by James Anderson of Arizona State University (http://www.epa.gov/Border2012/). The EFPM 2.5 must be quite low from our brick kilns as the sum of the species on the two kiln filters was 1.24 and 1.96 g/kg, respectively. Some of the particles being 20 produced in the fire-box may be deposited on the bricks and kiln walls. Despite the low particle emission factors for these kilns, brick making kilns are known to cause locally severe air quality impacts in Mexico as documented by Anderson, who reported PM 10 in homes and an elementary school near brick kilns well above 1000 µg/m 3 . Blackman et al. (2006) reported that the 330 brick making kilns in Ciudad Juarez (population 25 1.2 million) produced 16% of the PM and 43% of the SO 2 in the urban airshed. A large reduction in the total emissions from brick kilns is possible at the regional-national ACPD Introduction

Tables Figures
Back Close

Printer-friendly Version
Interactive Discussion scale by switching to more fuel efficient designs such as the vertical shaft brick kiln (http://www.vsbkindia.org/faq.htm). To our knowledge, there are no other published data on trace gas and particle emissions for brick making kilns that use wood or cellulose-based waste products as the primary fuel. An inventory of China's CO emissions was constructed following the Trans-5 port and Chemical Evolution over the Pacific (TRACE-P) campaign of 2001 (Streets et al., 2003). Those data were recently reevaluated to include a much larger contribution from coal-fired brick kilns (Streets et al., 2006). In a modeling study of aerosol over south Asia, a lack of seasonal variability for Kathmandu was credited to the exclusion of brick kiln emissions from the model (Adhikary et al., 2007). Nepalese kilns are also 10 fueled primarily by coal.
The impact of industrial biofuel use will likely remain difficult to assess for some time. The diverse range of micro-enterprise fuels (biomass, motor oil, tires, garbage, propane, coal, crop residues, etc.) makes it difficult to envision a tracer-based method that would quantitatively retrieve the contribution of this sector of the economy. Survey- 15 based methods, which likely work well for household biofuel use, may be less accurate when applied to highly competitive enterprises operating on thin margins. For example, in the report by Anderson cited above, stockpiled tires were a common sight at brick kilns. However, 100% of owners surveyed responded that they never burned tires while 12% responded that other kiln owners did.

Charcoal making kilns
The particle and trace gas emissions data for the charcoal making kilns are in Tables 4 and 6, respectively. As noted by Christian et al. (2007) the chemistry of the kiln emissions changes over the course of the approximately one-week production cycle. Specifically the molar ratio of total (measurable) VOC to CO increases by about a fac- 25 tor of 8 over this time. Thus, it is most meaningful to compare measurements from the same point in the production cycle, which we have done in Fig. 4 (Christian et al., 2007) and Africa . Table 6 also includes a simple comparison between the average of the charcoal making EF measured on days 3-5 in Mexico and the average EF for days 1-4 from a Zambian charcoal kiln . The Mexico MCE is slightly higher than the Zambian MCE, but the Mexican EF for acetic acid was 360% higher. The 5 higher acetic acid could be due to oak species being the primary fuel type in Mexico. Tannin content is high in oak species and, as polyphenols, tannins contain abundant hydroxyl and aectyl functional groups. As such they seem likely precursors for pyrolytic formation of acetic acid. Andreae and Merlet (2001) recommended an EFPM 2.5 for charcoal making of 4.0 g/kg. The sums of the EF for our PM 2.5 species in the two kiln filters were below that at 0.56 and 1.65 g/kg, respectively (Table 4). While a thick white smoke emanates from the vents in the kiln walls, much of the particulate matter produced inside probably remains embedded in the dirt walls while most of the gases are vented. 15 Our emissions data for crop residue fires are in Tables 4 and 5 in the columns labeled "stubble burns." Yokelson et al. (2009) reported airborne EF measurements for six crop residue fires in the Yucatan peninsula. Since airborne platforms tend to sample smoke with a higher flaming/smoldering ratio than ground-based platforms, we expect that the airborne samples will have a higher MCE and lower EF for smoldering compounds and 20 PM 2.5 . This pattern is observed. The airborne average MCE was 0.934, while the ground-based average MCE was 0.896. The airborne average EF for smoldering compounds (excluding formic acid) and PM 2.5 were 65±12% of the average EF measured from the ground. The formic acid average EF was higher for the fires sampled from the air, possibly due to fuel differences. The type of crop residue burning could not 25 be identified from the air. In the future, after analyzing additional airborne samples of crop residue fires from our flights in Mexico, we plan to recommend EF weighted by the relative fuel consumption for flaming and smoldering combustion.  Table 4 includes a very low EC/OC ratio for the one filter sample of stubble burning smoke. K and Na were very high on this filter and are known to catalytically lower the combustion temperature of black carbon during thermal evolution carbon analysis methods (Martins et al., 1998;Novakov and Corrigan, 1995). The high chloride content in the stubble burn filter may be linked to the use of agricultural chemicals. Typical 5 examples of these products include the herbicide 2,4-D dimethylamine salt (26.6% Cl by mass), the fungicide chlorthalonil (1,3-dicyanotetrachlorobenzene, 53% Cl), and the pesticide Lindane (or Kwell, hexachlorocyclohexane, 73% Cl). Two fires were sampled by the NCAR C-130 during MILAGRO in fuels that could not be identified from the air (fires #1 and #3 sampled on 23 March 2006, Yokelson et al., 2009). The particles 10 emitted by these fires were high in both Cl − and NO − 3 so they were probably crop residue fires. Fire plays a role in redistributing agricultural chemicals from their original point of application and they then build up in fire-free areas such as the Arctic (Becker et al., 2009).

15
In this section we discuss the impact of the sources sampled in this work at various scales. For cooking fires we start with national scale assessments for two different scenarios: a mostly rural developing country (Zambia) and a mostly urban developing country (Mexico).

developing country
Zambia has a total population of 10 million, of which 60% is rural. As much as one-third of the area of the country can burn in open fires in a single year which, combined with the small population, suggests that cooking fires would be much less important than open burning. However, standard estimation methods reveal otherwise. In Fig. 5   . EF for these combustion types, specific to Zambia when available, are from several sources (Andreae and Merlet, 2001;Bertschi et al., 2003;Christian et al., 2003;Sinha et al., 2004;Yokelson et al., 2008). We divide the annual biofuel emissions by two to estimate the dry season cooking fire emissions, and all the annual open burning emissions are assumed to be generated in the dry season. From Fig. 5 it is apparent that the dry season cooking fire emissions equal or exceed the dry season open burning emissions for four of the 17 species measured from both sources. For 11 of the 17 species, the dry season cooking fire emissions are at least 33% of the total dry season pyrogenic emissions. Thus, for those 11 species the annual emissions are estimated to be larger from cooking fires. 15 Mexico's total population of 100 million is 75% urban. Mexico City (population 20 million, ∼double the entire country of Zambia) is the second largest Megacity on earth and an acknowledged major source of pollutants. Mexico has experienced strong rural to urban migration and is considerably more developed than Zambia. We roughly estimate annual biofuel use in Mexico using three tables in Yevich and Logan (2003).

20
From their times more NMOC are generated from cooking fires than from urban areas. In addition, PM 2.5 is estimated to be almost 20 times greater from cooking fires than from urban areas on a national basis. Even if we allowed for a higher degree of secondary aerosol formation in fossil fuel emissions, which may not be the case, the cooking fires clearly dominate. If these estimates are right, then a switch to cleaner burning, more efficient 10 stoves (Sect. 3.1) could provide an enormous reduction of emissions on the national scale.
There is also a possibility that the amount of biofuel use in urban areas is underestimated. Marley et al. (2009) reported that 70% of the carbon in the ambient MCMA aerosol was modern and ascribed this to open biomass burning and garbage burn- 15 ing. However, garbage burning consumes a large fraction of plastics made from fossil fuels. If open burning was the dominant particle source in the MCMA and ventilation rates were similar year round, the PM 10 levels should peak in March-May when nearly all the open biomass burning occurs. Instead the PM 10 data show at best a weak increase in PM 10 during these months (Fig. 6) indicating that a different, year round 20 source of modern carbon could be "embedded" in the urban area. Two possibilities are cooking fires and industrial biofuel use.

Garbage burning impacts on the local-global atmosphere
We start this section by noting that the prevalence of open burning of garbage may be greater than commonly supposed even in developed countries. As noted earlier, 25 it has been estimated that 12-40% of rural households in the US burn garbage in their backyards (USEPA, 2006 Bates (2004) estimated that, at any one time, deep seated fires are occurring at about 80% of landfills. If we assume that 1000 Tg of garbage with a Cl content of 0.9% are burned each year globally (see introduction and Sect. 3.2), this could volatilize as much as 9 Tg/yr of Cl. This suggests a potential global source of 6-9 Tg/yr of HCl. By comparison, 5 the Reactive Chlorine Emissions Inventory (Keene et al., 1999) estimated that HCl from garbage burning was 2 Tg/yr and that sea salt dechlorination was the main global source at 7.6 Tg/yr. Recent HCl profiles in the marine boundary layer (Kim et al., 2008) may indicate that the sea salt dechlorination HCl source was over estimated. Our measurements indicate that the garbage burning HCl source may have been underes-10 timated. In general, Keene et al. (1999) found that additional HCl sources totaling to 42 Tg/yr were needed to balance the HCl budget. With the above in mind, we propose that garbage burning may be a considerably more important tropospheric source of HCl than previously assumed. We also note that many of the other main HCl sources, such as sea salt and volcanoes, can often be associated with a humid environment 15 and rapid removal of HCl (Tabazadeh and Turco, 1993). In dry environments, such as central Mexico where we measured water mixing ratios as low as 890 ppm, a larger fraction of freshly emitted HCl might react with OH to release Cl atoms. The latter would then react with NMOC. In any case, the HCl from garbage burning in dryer areas could have a longer lifetime and higher relative importance than the same amount 20 of HCl emitted in wetter areas.
We examined data obtained by other MILAGRO investigators for possible evidence of garbage burning. A particles-into-liquid-sampler (PILS) deployed by Georgia Tech at the MILAGRO T1 ground station north of Mexico City during March 2006 observed significant levels of chloride (up to 6 µg/m 3 ) for most of the month, with an average 25 of 0.5 µg/m 3 compared to 33 µg/m 3 total PM 2.5 (Greg Huey, personal communication, 2009). This translates to a mass ratio of 0.015. The average mass ratio of Cl − to the sum of particle species in our nascent smoke from garbage burning ( garbage burning could contribute about one-third of the PM 2.5 in the MC airshed. However, the PILS chloride may also reflect a contribution from agricultural fires and brick making kilns (Table 4) and volcanoes (e.g. Burton et al., 2007). We note that 3 of the 4 landfills we sampled are within ∼35 km to the west, north, and east of the T1 site (Table 1). We also note that EFCl − for brick kiln 1 was high and that this kiln is only ∼20 km west of T1. In addition, brick kiln 1 was one of many brick kilns in the region. We also looked for evidence of chlorine atom chemistry in the hydrocarbon ratios measured by whole air sampling. A plot of i-butane versus n-butane for 62 canister samples collected from both airborne and ground based sampling locations in and around MCMA gave an average i-butane/n-butane ratio of 0.33 (r 2 =1.00, Don Blake, 10 Barbara Barlett, personal communication, 2009). This is consistent with minimal chlorine atom oxidation of alkanes in the air sampled (Kim et al., 2008).
We make two other general points about garbage burning. More work is needed to measure other chlorinated emissions from burning refuse, including CH 3 Cl, which is also a proposed biomass burning tracer (Lobert et al., 1991). Secondly, PVC (the 15 primary source of HCl in garbage burning emissions) is also the most important predictor of dioxin emissions from the open burning of domestic waste (Neurath, 2004), so removing PVC from the waste before burning should have multiple benefits.

Conclusions
This work measured initial emission ratios and emission factors for trace gas and parti-20 cle species from five prominent, little-studied combustion sources: wood cooking fires, garbage burning, brick and charcoal making kilns, and crop residue burning. For Zambia, a mostly rural developing country, annual emissions from cooking fires likely exceed those from savanna fires for ∼11 of the 17 most abundant species measured from both sources. For Mexico, a mostly urban developing country, NMOC from cooking fires 25 are estimated to be substantially higher than from urban fossil fuel sources. Furthermore, cooking fires emit far more primary PM 2.5 than urban fossil fuel sources on a ACPD Introduction

Conclusions
References Tables  Figures   Back  Close Full Screen / Esc

Printer-friendly Version
Interactive Discussion national scale in Mexico. Therefore, the use of improved stoves that reduce both total fuel consumption per cooking task and the pollutant emissions per unit fuel consumption could provide large reductions in pollutant emissions throughout the developing world. Cooking fires produce far less HCN per unit mass of fuel than landscape-scale open biomass burning. Thus, the use of HCN ER or EF appropriate for open burning 5 to estimate regional biomass burning could cause a substantial underestimate of total biomass burning in areas where cooking fires are common (most of the developing world). Acetonitrile emissions from cooking fires are also likely to be much lower than for other biomass burning types. In addition, levoglucosan, K, and CH 3 Cl are likely not suitable as biomass burning tracers in areas where garbage burning is common.

10
Since cooking fires are also not detected from space a need exists to identify a chemical tracer for this source, which is currently quantified only by user surveys. Garbage burning is a globally significant source of particles and trace gases and may be a major global source of HCl. The emissions of particle EC, OC, levoglucosan, and K were similar for garbage burning and cooking fires. Of the three anhydrosugars 15 we measured, galactosan was the one most closely associated with biomass burning. Thus, it shows some promise for distinguishing between garbage burning and general biomass burning. Antimony (Sb) is a potential tracer for garbage burning. Comparison of both the Sb and the Cl − mass concentrations in the ambient PM 2.5 sampled at points in the MCMA airshed with the mass concentration of these species in PM 2.5 in fresh 20 garbage burning plumes tentatively suggests that landfill fires may produce about onethird of the PM 2.5 in some areas of the MCMA. Wood-fueled brick making kilns emitted a suite of trace gases similar to those from biomass burning, but with much lower emission factors. Smoke from these kilns had a very high EC/OC ratio. Charcoal making kilns in Mexico, Brazil, and Africa exhibited 25 a consistent pattern of increasing VOC/CO emission ratios over their approximately one week lifespan. The Mexican charcoal kilns produced higher acetic acid to CO ratios than an African kiln, possibly as a result of using tannin rich oak fuel. The PM 2.5 emission factor for both kiln types was evidently low relative to other biomass burning ACPD Introduction

Conclusions
References Tables  Figures   Back  Close Full Screen / Esc

Printer-friendly Version
Interactive Discussion types, possibly as a result of particle-scavenging on the kiln charge and walls. The fuels for brick making kilns vary substantially in ways that will likely remain difficult to quantify and limit the accuracy of regional air quality assessments and global emissions inventories. A single filter sample of smoke from a crop residue fire had very high levels of chloride that probably resulted from the use of agricultural chemicals.  Environ. Sci. Technol., 30, 104-109, 1996. Brocard, D., Lacaux, C., Lacaux, J. P., Kouadio, G., and Yoboué, V , 13, 542-545, 1963. Fernandes, S. D., Trautmann, N. M., Streets, D. G., Roden, C. A., and Bond, T. C.: Global 10 biofuel use, 1850-2000, Global Biogeochem. Cy., 21, GB2019, doi:10.1029/2006GB002836, 2007