Characterizing Biofuel Combustion with Patterns of ... - ACS Publications

Apr 25, 2012 - When a chimney is added, the stoves produce more black particles but also ..... Ryan J. Thompson, Jihua Li, Cheryl L. Weyant, Rufus Edw...
0 downloads 0 Views 2MB Size
Article pubs.acs.org/est

Characterizing Biofuel Combustion with Patterns of Real-Time Emission Data (PaRTED) Yanju Chen,† Christoph A. Roden,†,‡ and Tami C. Bond*,† †

Department of Civil and Environmental Engineering, University of Illinois at Urbana−Champaign, 205 N. Mathews Avenue, Urbana, Illinois 61801, United States S Supporting Information *

ABSTRACT: Emission properties and quantities from combustion sources can vary significantly during operation, and this characteristic variability is hidden in the traditional presentation of emission test averages. As a complement to the emission test averages, we introduce the notion of statistical pattern analysis to characterize temporal fluctuations in emissions, using cluster analysis and frequency plots. We demonstrate this approach by comparing emissions from traditional and improved wood-burning cookstoves under infield conditions, and also to contrast laboratory and in-field cookstove performance. Compared with traditional cookstoves, improved cookstoves eliminate emissions that occur at low combustion efficiency. For cookstoves where the only improvement is an insulated combustion chamber, this change results in emission of more light-absorbing (black) particles. When a chimney is added, the stoves produce more black particles but also have reduced emission factors. Laboratory tests give different results than in-field tests, because they fail to reproduce a significant fraction of low-efficiency events, spikes in particulate matter (PM) emissions, and less-absorbing particles. These conditions should be isolated and replicated in future laboratory testing protocols to ensure that stove designs are relevant to in-use operation.



INTRODUCTION In order to represent emissions from combustion sources in atmospheric models, emission quantities are needed; for particulate emissions, composition or absorption and scattering may be needed as well. However, these emission properties can vary among combustion units and operating conditions.1 Even if models require only average emission rates or properties from a population of sources, variability makes it difficult to gain confidence that emission tests are representative of the entire population. Ideally, many individual sources must be measured under a range of operating conditions to adequately capture emission properties. Such efforts have been used to characterize diesel engines2,3 and biomass burning, e.g., the Fire Laboratory at Missoula Experiments (FLAME)4−6 and have improved emission databases substantially. However, even a large sample might not isolate the reasons for variability. Variability in combustion emissions results from the geometry of the combustion appliance, fuel characteristics, air flow through combustion zones, and mixing of fuel and air. These factors can vary on characteristic time scales of several seconds, and spatial scales of a few millimeters. It is difficult to identify such varying factors in well-controlled settings and perhaps impossible for in-use combustion although it is theoretically possible. However, even if the controlling factors cannot be isolated, we suggest that they give rise to characteristic emission patterns that can be characterized © 2012 American Chemical Society

semiquantitatively using real-time measurements. In turn, this analysis relying on Patterns of Real-Time Emission Data (PaRTED) can identify the most common operating conditions, offering a complement to simple emission test averages, which do not convey the causes of variability. In the PaRTED approach, combustion is treated as a series of combustion “events” or short time segments. Simple measurements that can be obtained in real-timecombustion efficiency, particle properties, and proxies for emission quantitiesdescribe dominant combustion phases and their relative contributions to total emissions. These patterns can be used to evaluate performance similarities under a variety of conditions. In this study, we develop and demonstrate pattern analysis for emissions of residential biofuel combustion, a major source of primary carbonaceous aerosols. Globally, about 20% of black carbon (BC) and primary organic carbon (OC) are attributable to this source, and most of this biofuel is burned for cooking. Source variability means emission quantities and properties are uncertain,7−9 and it affects the accuracy of global and regional emission inventories used to study emission impacts on Received: Revised: Accepted: Published: 6110

January 26, 2012 April 22, 2012 April 25, 2012 April 25, 2012 dx.doi.org/10.1021/es3003348 | Environ. Sci. Technol. 2012, 46, 6110−6117

Environmental Science & Technology

Article

Table 1. List of Emission Variables variables

denotation and unit

measured variables scattering coefficient

bsp (Mm−1, 530 nm) bap_B (Mm−1, 467 nm) bap_G (Mm−1, 530 nm) bap_R (Mm−1, 660 nm) CO (ppm), CO2 (ppm)

absorption coefficient gaseous carbon concentration derived variablesa (1) indicator of combustion conditions modified combustion efficiency (2) optical characteristics single scattering albedo (530 nm) absorption Ångström exponent (3) emission quantities instantaneous scattering emission factor a

measurement or calculation measured by nephelometer (time resolution: 4 s) measured by PSAP (time resolution: 4 s) measured by gas sensors (time resolution: 1 s)

MCE

CO2/(CO + CO2)

SSA AAE

bsp/(bsp + bap_G) −ln(bap_G/bap_B)/ln(530/467)

IEFscat (m2/kg wood)

bsp/wood burntb

b

Measurement data were averaged over 2-min intervals. Wood burnt was calculated by carbon balance method,35 and details are described in the SI.

radiative transfer and long-range transport.10−13 In biofuel cookstoves, small-scale combustion variations are affected by macroscopic factors including ignition, fire feeding, wood size, fuels used for special purposes, or inclusion of waste fuels that are not part of the apparent regional resource base. User practices, while highly individual, may have regional averages governed by common customs, and a general pattern of emission may become apparent using the statistical approaches we describe. The purpose of this work is to formalize the PaRTED approach using a data set from one region. Comparisons among different regions will appear in subsequent work. We use the PaRTED approach to address two questions. First, why do emissions from different types of in-field cookstoves vary? Second, why do laboratory and in-field tests produce dissimilar results? The latter question is important for putting laboratory results into perspective. Measurements of biofuel emissions have frequently used laboratory standard tests to prescribe stove operation (e.g., refs 14−17), except for a few performed during simulated cooking18,19 and real cooking.9,20−22 Laboratory tests allow reproducible conditions, and provide environments where intensive emission properties can be measured. In contrast, in-field measurements are often conducted under challenging conditions, so that fewer emission properties can be obtained. However, current laboratory tests do not simulate realistic emission quantities and properties.9,23 Because of its repeatability, laboratory testing will remain the reference procedure to test stove performance. It is therefore essential to seek connections between laboratory and in-field tests, and to ensure that laboratory tests reproduce conditions in the field.

The samples then traveled through conductive tubing to the ARACHNE system. Particles larger than 4 μm were removed with a cyclone (URG, URG-2000−30EG). Real-time scattering and absorption coefficients of emitted particles were measured every four seconds with a single wavelength nephelometer (Radiance Research, M-903, 530 nm) and a 3-wavelength Particle Soot Absorption Photometer (PSAP, Radiance Research, 467 nm, 530 nm, 660 nm), respectively. Gaseous concentrations of CO (City Technology, 3ME/F) and CO2 (Telaire, 6004-S5000) were measured every second. Raw data from the nephelometer and PSAP were corrected for instrument artifacts.24,25 Scattering and absorption coefficient, CO and CO2 concentration were converted to standard conditions (1 atm and 20 °C) and background readings were subtracted. In-field tests measured emissions from traditional and improved stoves in and around Suyapa, Honduras during the summers of 2004, 2005, and 2006. Traditional mud or clay cookstoves (TRAD) varied in size and shape, with no standard fuel opening. Improved stoves were designed to improve efficiency by adding an insulated combustion chamber either with a chimney (ICCh) or without (ICN), although design recommendations were not always implemented by the manufacturer. Laboratory tests were conducted at Aprovecho Research Center in 2005, 2006, and 2007, by completing a standard water-boiling test (WBT). Only ICN cookstoves were tested in the laboratory, as the emission hood there was not designed for chimney stoves. Our sample is similar in size or larger than those in previous studies of cookstove emissions and the variability of our test results are comparable with other stove tests. Further description of in-field and laboratory tests and results can be found in Roden et al.20,23 The present work uses the data from these studies to demonstrate the PaRTED approach. Data Reduction and Processing. In this study, a combustion “event” is defined as a two-minute segment. The optimal averaging period captures major fluctuations while eliminating noise, and was investigated using Fourier analyses (see SI). Several emission properties were derived from measured variables as summarized in Table 1. Only normalized quantities, which are largely independent of air parcel dilution, are used for PaRTED.



MATERIALS AND METHODS Measurement Overview. Our procedure and instrumentation for measuring emissions from biofuel combustion have been previously reported.20 Emissions were measured with a self-designed real-time sampling systemthe ARACHNE (Ambulatory Real-time Analyzer for Climate and Healthrelated Noxious Emissions). Smoke plume from cookstove combustion was sampled with an eight-armed multipoint stainless-steel probe. The probe was 1 m in diameter, which covered most of the emission plume. The probe was suspended 1−1.5 m above the cookstove, so the smoke plume was naturally diluted by the ambient air during the initial 1−1.5 s. 6111

dx.doi.org/10.1021/es3003348 | Environ. Sci. Technol. 2012, 46, 6110−6117

Environmental Science & Technology

Article

While clustering identifies properties that occur together, one-variable and two-variable frequency plots give a better visual representation of the combustion events. Unlike clustering results, frequency plots in this paper are weighted by particulate emission during each event, as represented by scattering, to show the conditions under which most of the particulate matter is emitted. Therefore, the term “emissionweighted” (EW) frequency plot will be used. Two-variable (joint) EW-frequency plots show which emissions occur most frequently together by classifying combustion events into bins (e.g., SSA and MCE). When these weighted frequencies are scaled by total emission rate, they give semiquantitative emission rates of each particle type. Here, the treatment of the data presented by Roden et al. 20 has been further developed to quantify emissions under different conditions. To summarize, cluster analysis is useful to identify major covariations among variables. Frequency plots show correlations visually and allow more quantitative comparison. The use of these two methods will be illustrated below.

Modified combustion efficiency (MCE) is defined as the molar ratio of emitted CO2 to the sum of CO and CO2.26,27 A value of 1 indicates complete combustion. In other literature, MCE is used to classify combustion into flaming or smoldering.1 We avoid definitive use of those terms here as during solid-fuel combustion a “devolatilization” phase also occurs, when pyrolysis releases organic material and gases, but the environment is not hot enough to ignite the fuel.28−30 Literature on open biomass burning indicates that particle emission factors increase as MCE decreases.1 However, this finding has not been evaluated for biofuel combustion, or for real-time measurements from any source. Biofuel particulate emissions consist mainly of BC and OC,18 which have different optical characteristics. Single scattering albedo (SSA) is the fraction of total extinction attributable to scattering (about 0.25 for the most absorbing aerosol, and 1.0 for purely scattering particles). Absorption Angstrom exponent (AAE) conveys the spectral dependence of absorption. BC is produced only in flames, and has SSA of 0.20−0.25 and AAE around 1.31 OC is the liquid product of pyrolysis that has not passed through a flame, and has much less absorption (higher SSA) and higher AAE (around 6−7 in extracts32−34). In this study, we use light scattering as a proxy for particulate mass emission quantities by assuming values of mass scattering cross-section (m2/g). Although these values vary with particle size and type, scattering can still demonstrate when mass emission rates are large, even if the inferred mass emissions are not exact. No real-time measurements of mass are free of artifacts, especially for semivolatile wood smoke; given a field of imperfect candidates, light scattering is a reasonable mass proxy that can be measured with low power requirements, as is required for in-field measurements. We define an instantaneous scattering emission factor (IEFscat) to convey the magnitude of particulate emission in near-real-time. IEFscat is the total cross-section (m2) of light scattering attributable to the particulate emissions from one kilogram of wood and therefore has units of m2/(kg wood). Calculation follows the carbon-balance method 35 (see SI). Test-averaged measurements with filter-measured mass show a relatively consistent relationship between light scattering and PM mass emission (Figure S3 of the SI). The average ratio between scattering emission and filter-based mass emission was 3.4 m2/g, so 10 m2/(kg wood) is approximately 3 g PM/(kg wood), although there is variability in this relationship. Statistical Analyses. Cluster analysis and frequency distribution were used to classify combustion events and their contributions to total emissions. We treated each combustion event as an object having a location in multidimensional space. K-means clustering 36 with a default squared Euclidean distance was used to find a partitioning in which objects within each cluster are as close to each other as possible, and as far from objects in other clusters as possible. The clustering variables were SSA, AAE, MCE, and IEFscat; other combinations of measurements are possible but are not independent, as shown in Table 1. The silhouette validation method37 was used to optimize the number of clusters. Silhouette values range from −1 to 1 for each point, with a value of 1 indicating that a point is optimally assigned to a cluster, and zero indicating that it could equally be assigned to another cluster. Pairwise comparison (Tukey−Kramer method) also checked whether properties were significantly different among clusters. Clustering results will be presented as a number frequency of events, demonstrating which types of events occur most often.



RESULTS AND DISCUSSION Overview of In-Field Test Data. Cluster analysis of infield combustion events identified three major groups of events, summarized in Figure 1 (numeric results in SI, Table S1).

Figure 1. Box plots showing medians and distributions of particle and combustion properties of three clusters for all in-field test data: lowSSA (BC-like), high-SSA (OC-like), and intermediate.

Particles emitted during events in Cluster 1 have a mean SSA of 0.28 and AAE of 1.38, similar to properties of BC that strongly absorbs light. These particles are emitted at relatively high MCE, averaging 0.95. Particles emitted during events in Cluster 3 have a mean SSA of 0.78 and AAE of 3.47; these particles are moderately absorbing and the absorption has a strong wavelength dependence, similar to properties of OC. These emissions occur at lower MCE averaging 0.84. The emission properties of Cluster 2 lie between the first two clusters. Since SSA is a distinctive property separating the clusters, we will refer to the clusters as “low-SSA” (BC-like, Cluster 1), “highSSA” (OC-like, Cluster 3), or intermediate (Cluster 2). These data cannot indicate whether Cluster 2 contains homogeneous 6112

dx.doi.org/10.1021/es3003348 | Environ. Sci. Technol. 2012, 46, 6110−6117

Environmental Science & Technology

Article

Figure 2. PaRTED plot describing EW-frequency distribution of emission with respect to SSA and MCE: (a) one-variable frequency plot with respect to SSA; (b) two variable (joint) frequency plot; (c) one-variable frequency plot with respect to MCE. In the joint frequency plot, darker color indicates higher contribution to total emission, and the magnitude of the contribution (%) is shown in the color bar. Total emission factor can be obtained as the sum of emission factor values (secondary axes) in the one-variable plots. Part (d): assignment to clusters in Figure 1 is provided for comparison.

frequency distributions). When this plot is scaled by total emission rate, the area under the curve indicates the magnitude of PM emission at a particular condition, subject to the approximate relationship between light scattering and mass. For the in-field data, the largest peak in the one-variable plot for SSA (Figure 2a) occurs around SSA = 0.3 (cluster 1), again indicating that black particles produced at high MCE are a large fraction of the PM. Smaller peaks also appear at intermediate and high SSA (clusters 2 and 3), and the scattered contribution in the joint plot is apparent from the broadness of the peaks. In the one-variable plot by MCE (Figure 2c), a slightly distinguishable peak appears around MCE = 0.97, mainly from points assigned to the low-SSA cluster. The data show a broad distribution of MCE with about half the emissions occurring above 0.90 and a tail extending down to MCE of 0.60. The lower-MCE events include contributions from both high-SSA and intermediate clusters. Other joint EW-frequency plots (Figure S7 of the SI) show that low-SSA particles have AAE around 1 while high-SSA particles have AAE greater than 3. Most of the particle emission occurs in events with IEFscat less than 100 m2/(kg wood), although there is a substantial contribution from events with higher emissions, which are slightly more common in high-SSA events. Peaks in the onevariable plot may shift by 1−2 bins depending on the averaging times of the real-time data (see SI). The uncertainty will not change our conclusions based on the one-variable plots. Comparison among Three Cookstove Types. Averaged properties for the three cookstove types are shown in Table 2. Particles emitted from TRAD stoves have higher SSA and AAE and lower MCE than improved stoves. EFscat is highest for ICN stoves followed by TRAD and then ICCh stoves, consistent with PM emission factors reported by Roden et al.23 We now use both cluster analysis and frequency plots to examine the emission variation among stove types. Table 2

particles with intermediate properties or a heterogeneous combination of Cluster 1 and Cluster 3 particles. Overall, 44% of combustion events produce particles in the low-SSA cluster, 24% have properties in the high-SSA cluster and 32% have intermediate properties. Most instantaneous emission magnitudes (IEFscat) lie between 0 to 100 m2/(kg wood). The difference in IEFscat among the three clusters is not as striking as the other parameters, although it is statistically significant at a level of 0.05. On average, IEFscat is greatest for the high-SSA cluster, followed by intermediate and low-SSA. However, highemission events occur in all three clusters. The averaged silhouette value was greatest for three clusters, rather than two or four. Ninety-six percent of the data points had silhouette values greater than zero and the mean silhouette value was 0.51, meaning the averaged distance of points among clusters is twice of that within the cluster. Pairwise comparisons (Tukey-Kramer method) showed that SSA, AAE, MCE, and IEFscat were different among clusters at a significance level of 0.05. Figure 2 shows EW-frequency plots for SSA and MCE for infield data of all stoves. The joint EW-frequency plot (Figure 2b) shows the covariance of the two properties. Because frequencies are weighted by PM emission rate, they indicate the fractional contribution to emission for each type of event. Figure 2d indicates the cluster to which each event belongs (see Figure 1). Substantial contribution to total emission comes from cluster 1dark particles emitted at relatively high efficiency, with SSA from 0.2 to 0.4 and MCE from 0.92 to 1. Contributions from intermediate and high-SSA clusters are evident, but not as concentrated around single properties as those in the low-SSA cluster. Although the joint EW-frequency plot is the most visually striking, one-variable EW-frequency plots are easier to interpret semiquantitatively (Figure 2a,c). These plots are the integrals of the joint EW-frequency plot over one variable (marginal EW6113

dx.doi.org/10.1021/es3003348 | Environ. Sci. Technol. 2012, 46, 6110−6117

Environmental Science & Technology

Article

emitting events. The general trend in Figures 3 and 4 again shows that improving stoves from TRAD to ICN to ICCh increases MCE and reduces high-SSA clusters. High-emission events with IEFscat greater than 100 m2/(kg wood) occur in all three types of stoves. These are often associated with high SSA (Figure S8 of the SI) and low EF, meaning most of the absolute PM emission occurs at events with relatively low IEFscat. However, these events form a large fraction of the total emission for ICN stoves (green peaks at 100 and 150 m2/kg wood in Figure 4c), with the result that emissions from ICN stoves are slightly higher than TRAD stoves. Thus, both ICN and ICCh have similar number frequencies of events (Table 2), but the occurrence of highemission events makes total emission from ICN stoves twice that from ICCh stoves. This fact is more apparent in Figure 5, which shows the cumulative emission rate for IEFscat (the integral of Figure 4c). ICN and ICCh curves are similar until IEFscat is greater than 50 m2/kg wood, indicating that the two stove types operate at similar emission rates under lowemission conditions, but that higher-emission events result in much higher overall emissions for the ICN stoves. Two-thirds of the gap between ICN and ICCh comes from emission events above 100 m2/(kg wood). Comparison between in-Field and Laboratory Tests. Particle emission factors are 68% lower for laboratory tests compared with in-field tests of the same stoves (Table 2). This difference is confirmed by gravimetric emission factors for the same tests.23 Here, we explore this difference using the PaRTED approach. This discussion focuses only on ICN stoves, which were tested in the laboratory. Figure 3 shows that events in laboratory tests are more tightly clustered around single properties compared with infield measurements, indicating the combustion was more consistent. By number frequency, most laboratory events (83%) fall into the low-SSA cluster, with 17% in the intermediate cluster and none at high SSA. This dominance of high-MCE, low-SSA events is apparent in joint EWfrequency plots (Figure 3). In contrast, in-field ICN tests show only 50% of combustion events in the low-SSA cluster, 33% in the intermediate cluster, and 17% in the high-SSA cluster (see SI, Table S3).

Table 2. Summary of in-Field and Laboratory Tests, Including General Testing Information, Averaged Emission Properties and Results from Clustering Analysis in-field general testing information stove type no. of tests no. of events averaged properties EFscat (m2/kg) single scattering albedo (SSA) absorption Ångström exponent (AAE) modified combustion efficiency (MCE) percent events in each cluster low-SSA intermediate high-SSA

laboratory

TRAD 10 1079

ICN 25 1461

ICCh 15 691

ICN 8 203

21.4 0.52 1.80

23.6 0.46 1.79

13.6 0.42 1.42

7.6 0.32 1.29

0.89

0.92

0.93

0.95

29% 30% 41%

58% 29% 13%

50% 34% 16%

83% 17% N/A

shows the number frequency of combustion events in each cluster for three stove types. TRAD stoves have a large fraction of high-SSA events (41%) and this fraction decreases to 13% and 16% for ICN and ICCh stoves, respectively. Reduction of high-SSA, low-MCE events is consistent with design of the improved cookstoves to increase combustion efficiency. The removal of these high-SSA events means that test-averaged SSA and AAE for these stoves are lower than for traditional stoves. Stove improvements increase efficiency, decrease high-SSA emissions, and therefore produce a greater fraction of BC-like particles. Figure 3 summarizes the joint EW-frequency plots of SSA and MCE for the three cookstove types. Figure 4 shows onevariable EW-frequency plots of SSA, MCE, and IEFscat. Curves for each stove are scaled to the proxy for total PM emissions (scattering) to provide absolute, rather than relative, comparisons among stove types. For TRAD stoves, PM emissions have SSA from 0.2 and 1.0, with peaks at low, intermediate, high SSA. Most PM (96%) is emitted from MCE = 0.8−1. For ICN stoves, emission from low-MCE, high-SSA events is slightly reduced (see also Table 2). ICCh stoves clearly increase MCE, reduce high-SSA (OC-like) events, and have fewer high-

Figure 3. Joint EW-frequency plots of three cookstove types from in-field tests and cookstoves tested in the laboratory with respect to SSA and MCE. Cookstove type in the field: (a) traditional (TRAD), (b) isolated chamber without chimney (ICN), (c) isolated chamber with chimney (ICCh). Only ICN stoves were tested in the laboratory (d). Color bar indicates contribution of events to total PM emission for each test type. 6114

dx.doi.org/10.1021/es3003348 | Environ. Sci. Technol. 2012, 46, 6110−6117

Environmental Science & Technology

Article

Figure 4. Distribution of emission factors with respect to (a) SSA, (b) MCE, and (c) IEFscat for three stove types. Particles with different properties are classified into bins. The height of the curve represents the amount of scattering, a proxy for mass, contributed by the particles in each bin. Total scattering emission factor can be obtained by summing over the entire curve. The width of each bin is (a) ΔSSA = 0.02; (b) ΔMCE = 0.01; and (c) ΔIEFscat = 10 m2/kg wood.

these tests to evaluate performance will not be optimized for real operation. Johnson et al.9 suggested that laboratory test protocols should be based on a distribution of observed, realtime MCE. We further propose that protocols should reproduce the factors leading to emission of different particle types, which can be observed in real time. This is especially important because much evidence suggests that the primary adverse health impacts of solid-fuel cooking occur from particulate matter.38,39 We suggest that in future laboratory tests, a regional pattern of operation should be developed by comparing PaRTED analyses of traditional stoves in the laboratory and field. Low-MCE, high-SSA events should be produced, first by exploring the most likely differences between laboratory and field: fuel quality, size, and loading practice. If patterns similar to field operation cannot be generated in laboratory testing, then additional observations need to be conducted in field settings; for example, real-time cameras synchronized with the data collection. After operating procedures for traditional stoves have been shown to reproduce PaRTED analyses in laboratory settings, the same stove operating procedures should be used when testing improved stoves. The in-field results presented here were taken in only one location, and findings may vary for cookstoves tested in other regions. This study developed the approach of characterizing emissions from biofuel combustion using time-resolved emissions, cluster analysis, and graphical representations of EW-frequency. Such a presentation captures operational characteristics that cannot be obtained from test averages. In future work, we will use these tools to examine statistical similarity among regions and conditions.

Figure 5. Cumulative emission factors with respect to IEFscat for stoves tested in the field and laboratory. The cumulative emission factor at each value of IEFscat represents the total emission from events with IEFscat at or below that value.

Figure 5 shows that in laboratory tests, IEFscat never exceeds 50 m2/(kg wood) (also see Figure S8 of the SI). Laboratory testing always operates in the low-emission mode, which produces only 30% of the total particulate emission observed in the in-field tests. Laboratory tests of ICN stoves do not reproduce high-emission events. The tests are dominated by events with high MCE that produce BC-like particles, and may miss the high-SSA OC-like emission observed in the field tests. Implications. The generation of cookstoves tested here focused only on improving efficiency by insulating combustion chambers and adding a chimney. ICN stoves begin to eliminate some of the emissions at low MCE, but in the stoves observed here, overall PM emissions are not substantially reduced due to high-emission events. ICCh stoves reduce PM emission per fuel by eliminating high-SSA events with low MCE, so that emissions contain a greater fraction of black particles. ICN stoves, however, may have improved heat transfer efficiency, thereby reducing total fuel use and total emissions. This aspect of stove improvement was not explored here. Laboratory tests of ICN stoves compare poorly with field tests, because they fail to reproduce a significant fraction of the events that have low MCE, high-SSA particles and spikes in PM emission. Until the factors leading to these conditions are included in test protocols, laboratory tests will not accurately represent in-field emissions. Furthermore, designs that use



ASSOCIATED CONTENT

S Supporting Information *

Instrumental calibration, optimal averaging period, calculation of IEFscat using the carbon balance method, uncertainty analysis of data averaging, EW-frequency plots of SSA and AAE, SSA and IEFscat, and numeric results of cluster analysis for in-field and laboratory data. This material is available free of charge via the Internet at http://pubs.acs.org.



AUTHOR INFORMATION

Corresponding Author

*Phone: +1 (217) 244 5277; e-mail: [email protected]. 6115

dx.doi.org/10.1021/es3003348 | Environ. Sci. Technol. 2012, 46, 6110−6117

Environmental Science & Technology

Article

Present Address

domestic combustion of selected fuels. Environ. Sci. Technol. 1999, 33 (16), 2703−2709. (15) Oanh, N. T. K.; Albina, D. O.; Ping, L.; Wang, X. K. Emission of particulate matter and polycyclic aromatic hydrocarbons from select cookstove-fuel systems in Asia. Biomass Bioenerg. 2005, 28 (6), 579− 590. (16) Zhang, J.; Smith, K. R.; Ma, Y.; Ye, S.; Jiang, F.; Qi, W.; Liu, P.; Khalil, M. A. K.; Rasmussen, R. A.; Thorneloe, S. A. Greenhouse gases and other airborne pollutants from household stoves in China: A database for emission factors. Atmos. Environ. 2000, 34 (26), 4537− 4549. (17) Venkataraman, C.; Rao, G. U. M. Emission factors of carbon monoxide and size-resolved aerosols from biofuel combustion. Environ. Sci. Technol. 2001, 35 (10), 2100−2107. (18) Venkataraman, C.; Habib, G.; Eiguren-Fernandez, A.; Miguel, A. H.; Friedlander, S. K. Residential biofuels in south Asia: Carbonaceous aerosol emissions and climate impacts. Science 2005, 307 (5714), 1454−1456. (19) Habib, G.; Venkataraman, C.; Bond, T. C.; Schauer, J. J. Chemical, microphysical and optical properties of primary particles from the combustion of biomass fuels. Environ. Sci. Technol. 2008, 42 (23), 8829−8834. (20) Roden, C. A.; Bond, T. C.; Conway, S.; Benjamin, A.; Pinel, O. Emission factors and real-time optical properties of particles emitted from traditional wood burning cookstoves. Environ. Sci. Technol. 2006, 40 (21), 6750−6757. (21) Johnson, M.; Edwards, R.; Frenk, C. A.; Masera, O. In-field greenhouse gas emissions from cookstoves in rural Mexican households. Atmos. Environ. 2008, 42 (6), 1206−1222. (22) Brocard, D.; Lacaux, C.; Lacaux, J. P.; Kouadio, G.; Yoboue, V., Emissions from the combustion of biofuels in western Africa. In Biomass Burning and Global Change; Levine, J. S., Ed.; MIT Press: Cambridge, MA, 1996; pp 350−360. (23) Roden, C. A.; Bond, T. C.; Conway, S.; Pinel, A. B. S.; MacCarty, N.; Still, D. Laboratory and field investigations of particulate and carbon monoxide emissions from traditional and improved cookstoves. Atmos. Environ. 2009, 43 (6), 1170−1181. (24) Anderson, T. L.; Ogren, J. A. Determining aerosol radiative properties using the TSI 3563 integrating nephelometer. Aerosol Sci. Technol. 1998, 29 (1), 57−69. (25) Bond, T. C.; Anderson, T. L.; Campbell, D. Calibration and intercomparison of filter-based measurements of visible light absorption by aerosols. Aerosol Sci. Technol. 1999, 30 (6), 582−600. (26) Hao, W. M.; Ward, D. E. Methane production from global biomass burning. J. Geophys. Res. 1993, 98 (D11), 20657−20661. (27) Yokelson, R. J.; Griffith, D. W. T.; Ward, D. E. Open-path Fourier transform infrared studies of large-scale laboratory biomass fires. J. Geophys. Res. 1996, 101 (D15), 21067−21080. (28) Svoboda, K.; Hartman, M.; Cermák , J., Combustion MechanismsSolid Phase. In Pollutants from Combustion; Vovelle, C., Ed.; Springer: Netherlands: 2000; Vol. 547, pp 35−50. (29) Gelencser, A. Carbonaceous Aerosol; Springer: Dordrecht, The Netherlands, 2004. (30) Kozinski, J. A.; Saade, R. Effect of biomass burning on the formation of soot particles and heavy hydrocarbons. An experimental study. Fuel 1998, 77 (4), 225−237. (31) Bond, T. C.; Bergstrom, R. W. Light absorption by carbonaceous particles: An investigative review. Aerosol Sci. Technol. 2006, 40 (1), 27−67. (32) Kirchstetter, T. W.; Novakov, T.; Hobbs, P. V., Evidence that the spectral dependence of light absorption by aerosols is affected by organic carbon. J. Geophys. Res. 2004, 109, (D21). (33) Hoffer, A.; Gelencser, A.; Guyon, P.; Kiss, G.; Schmid, O.; Frank, G. P.; Artaxo, P.; Andreae, M. O. Optical properties of humiclike substances (HULIS) in biomass-burning aerosols. Atmos. Chem. Phys. 2006, 6, 3563−3570. (34) Chen, Y.; Bond, T. C. Light absorption by organic carbon from wood combustion. Atmos. Chem. Phys. 2010, 10 (4), 1773−1787.



Now at SPEC Inc., 3022 Sterling Circle, Suite 200, Boulder, Colorado, 80301 United States of America. Notes

The authors declare no competing financial interest.



ACKNOWLEDGMENTS Field tests were completed with the cooperation of Anibal Benjamin Osorto Pinel, Wendy Naira, and Augusto Ramirez at AHDESA and Stuart Conway at Trees, Water, and People. Laboratory tests were completed with the collaboration of Nordica MacCarty and Aprovecho Research Center. This project was supported by the National Science Foundation’s Atmospheric Chemistry Program under Grant No. ATM0349292, the U.S. Environmental Protection Agency’s Partnership for Clean Indoor Air, and the Clean Air Task Force.



REFERENCES

(1) Reid, J. S.; Koppmann, R.; Eck, T. F.; Eleuterio, D. P. A review of biomass burning emissions part II: intensive physical properties of biomass burning particles. Atmos. Chem. Phys. 2005, 5, 799−825. (2) Yanowitz, J.; McCormick, R. L.; Graboski, M. S. In-use emissions from heavy-duty diesel vehicles. Environ. Sci. Technol. 2000, 34 (5), 729−740. (3) Maricq, M. M. Chemical characterization of particulate emissions from diesel engines: A review. J. Aerosol Sci. 2007, 38 (11), 1079− 1118. (4) Chen, L. W. A.; Moosmüller, H.; Arnott, W. P.; Chow, J. C.; Watson, J. G.; Susott, R. A.; Babbitt, R. E.; Wold, C. E.; Lincoln, E. N.; Hao, W. M. Emissions from laboratory combustion of wildland fuels: Emission factors and source profiles. Environ. Sci. Technol. 2007, 41 (12), 4317−4325. (5) Chen, L. W. A.; Moosmüller, H.; Arnott, W. P.; Chow, J. C.; Watson, J. G.; Susott, R. A.; Babbitt, R. E.; Wold, C. E.; Lincoln, E. N.; Hao, W. M. Particle emissions from laboratory combustion of wildland fuels: In situ optical and mass measurements. Geophys. Res. Lett. 2006, 33 (4), L04803. (6) McMeeking, G. R.; Kreidenweis, S. M.; Baker, S.; Carrico, C. M.; Chow, J. C.; Collett, J. L.; Hao, W. M.; Holden, A. S.; Kirchstetter, T. W.; Malm, W. C.; Moosmuller, H.; Sullivan, A. P.; Wold, C. E. Emissions of trace gases and aerosols during the open combustion of biomass in the laboratory. J. Geophys. Res. 2009, 114, 20. (7) Butcher, S. S.; Sorenson, E. M. Study of wood stove particulate emissions. JAPCA J. Air. Waste Manage. Assoc. 1979, 29 (7), 724−728. (8) McCrillis, R. C.; Watts, R. R.; Warren, S. H. Effects of operating variables on PAH emissions and mutagenicity of emissions from woodstoves. J. Air Waste Manage. Assoc. 1992, 42 (5), 691−694. (9) Johnson, M.; Edwards, R.; Berrueta, V.; Masera, O. New approaches to performance testing of improved cookstoves. Environ. Sci. Technol. 2010, 44 (1), 368−374. (10) Bond, T. C.; Streets, D. G.; Yarber, K. F.; Nelson, S. M.; Woo, J. H.; Klimont, Z., A technology-based global inventory of black and organic carbon emissions from combustion. J. Geophys. Res. 2004, 109, (D14). (11) Streets, D. G.; Bond, T. C.; Carmichael, G. R.; Fernandes, S. D.; Fu, Q.; He, D.; Klimont, Z.; Nelson, S. M.; Tsai, N. Y.; Wang, M. Q.; Woo, J. H.; Yarber, K. F. An inventory of gaseous and primary aerosol emissions in Asia in the year 2000. J. Geophys. Res. 2003, 108 (D21), 8809. (12) Reddy, M. S.; Venkataraman, C. Inventory of aerosol and sulphur dioxide emissions from India. Part IIbiomass combustion. Atmos. Environ. 2002, 36 (4), 699−712. (13) Aunan, K.; Berntsen, T. K.; Myhre, G.; Rypdal, K.; Streets, D. G.; Woo, J. H.; Smith, K. R. Radiative forcing from household fuel burning in Asia. Atmos. Environ. 2009, 43 (35), 5674−5681. (14) Oanh, N. T. K.; Reutergardh, L. B.; Dung, N. T. Emission of polycyclic aromatic hydrocarbons and particulate matter from 6116

dx.doi.org/10.1021/es3003348 | Environ. Sci. Technol. 2012, 46, 6110−6117

Environmental Science & Technology

Article

(35) Smith, K. R.; Khalil, M. A. K.; Rasmussen, R. A.; Thorneloe, S. A.; Manegdeg, F.; Apte, M. Greenhouse gases from biomass and fossilfuel stoves in developing-countriesA Manila pilot-study. Chemosphere 1993, 26 (1−4), 479−505. (36) MacQueen, J. In Some Methods for Classification and Analysis of Multivariate Observations, Proceedings of the Fifth Symposium on Math, Statistics, and Probability; University of California Press: Berkeley, CA, 1967; pp 281−297. (37) Rousseeuw, P. J. SilhouettesA graphical aid to the interpretation and validation of cluster-analysis. J. Comput. Appl. Math. 1987, 20, 53−65. (38) Bolling, A. K.; Pagels, J.; Yttri, K. E.; Barregard, L.; Sallsten, G.; Schwarze, P. E.; Boman, C., Health effects of residential wood smoke particles: the importance of combustion conditions and physicochemical particle properties. Part. Fibre Toxicol. 2009, 6. (39) Naeher, L. P.; Brauer, M.; Lipsett, M.; Zelikoff, J. T.; Simpson, C. D.; Koenig, J. Q.; Smith, K. R. Woodsmoke health effects: A review. Inhal. Toxicol. 2007, 19 (1), 67−106.

6117

dx.doi.org/10.1021/es3003348 | Environ. Sci. Technol. 2012, 46, 6110−6117