This study focuses on improving source apportionment of carbonaceous aerosol in South Asia and consists of three parts: (1) development of novel molecular marker–based profiles for real-world biofuel combustion, (2) application of these profiles to a year-long data set, and (3) evaluation of profiles by an in-depth sensitivity analysis. Emissions profiles for biomass fuels were developed through source testing of a residential stove commonly used in South Asia. Wood fuels were combusted at high and low rates, which corresponded to source profiles high in organic carbon (OC) or high in elemental carbon (EC), respectively. Crop wastes common to the region, including rice straw, mustard stalk, jute stalk, soybean stalk, and animal residue burnings, were also characterized. Biofuel profiles were used in a source apportionment study of OC and EC in Godavari, Nepal. This site is located in the foothills of the Himalayas and was selected for its well-mixed and regionally impacted air masses. At Godavari, daily samples of fine particulate matter (PM2.5) were collected throughout the year of 2006, and the annual trends in particulate mass, OC, and EC followed the occurrence of a regional haze in South Asia. Maximum concentrations occurred during the dry winter season and minimum concentrations occurred during the summer monsoon season. Specific organic compounds unique to aerosol sources, molecular markers, were measured in monthly composite samples. These markers implicated motor vehicles, coal combustion, biomass burning, cow dung burning, vegetative detritus, and secondary organic aerosol as sources of carbonaceous aerosol. A molecular marker–based chemical mass balance (CMB) model provided a quantitative assessment of primary source contributions to carbonaceous aerosol. The new profiles were compared to widely used biomass burning profiles from the literature in a sensitivity analysis. This analysis indicated a high degree of stability in estimates of source contributions to OC when different biomass profiles were used. The majority of OC was unapportioned to primary sources and was estimated to be of secondary origin, while biomass combustion was the next-largest source of OC. The CMB apportionment of EC to primary sources was unstable due to the diversity of biomass burning conditions in the region. The model results suggested that biomass burning and fossil fuel were important contributors to EC, but could not reconcile their relative contributions.
 Atmospheric aerosols reflect or absorb sunlight and directly impact climate by changing the amount of solar radiation reaching the Earth's surface. Aerosols also indirectly impact climate by causing changes in cloud composition and physics by forming smaller, yet more numerous, cloud droplets, resulting in longer cloud lifetimes and enhanced cloud albedo [Ramanathan et al., 2005; Sun and Ariya, 2006]. In Asia, heavily polluted atmospheres have been observed in recent decades and were associated with long-term dimming over the continent [Ramanathan et al., 2005]. Long-term climatic impacts of the haze may also include changes to hydrology and agriculture [Centers for Cloud, Chemistry, and Climate, 2002]. The impacts of atmospheric aerosols on climate relate to their chemical composition: light absorption is dominated by dark-colored particles like elemental carbon (EC), also called black carbon or soot, whereas organic carbon (OC) and sulfate aerosols are highly reflective [Ramanathan et al., 2001].
 The composition of aerosols at a point location is a function of aerosol sources and their strengths. Chemical signatures can be used to identify aerosol sources and to quantitatively understand relative source contributions. Molecular markers are organic species present in the ambient atmosphere that come from specific aerosol source categories, are stable in the atmosphere, and can be used quantitatively in source apportionment [Schauer and Cass, 2000]. Chemical mass balance (CMB) modeling apportions carbonaceous aerosol to primary sources using chemically defined source profiles [Watson et al., 1984]. The CMB model has been used in studies of ambient aerosol worldwide [Chowdhury et al., 2007; Schauer et al., 2002; Stone et al., 2008; Zheng et al., 2006] and its results have been used in air resources management to understand pollution sources [Watson et al., 2008].
 Effective receptor-based source apportionment requires source profiles that are representative of the emissions impacting the receptor. In the case of biomass burning, the type of fuel burned and combustion conditions affect the chemical composition of particulate matter generated [Simoneit, 2002]. This presents a significant challenge in the selection of biomass source profiles since fuel type and burn conditions can vary considerably by region. Previous studies have characterized the organic composition of particulate matter generated by fireplace combustion of oak and pine [Schauer et al., 2001], modern-stove combustion of major tree species found in the United States [Fine et al., 2004a], wood-stove emissions of common Austrian leaves [Schmidl et al., 2008a] and fuelwoods [Schmidl et al., 2008b], modern-stove combustion of biomasses indigenous to Bangladesh [Sheesley et al., 2003], open burning of a pine forest [Lee et al., 2005], stove burning of straw [Zhang et al., 2007], and others. These studies have demonstrated that biomass emissions have chemical signatures distinct from other aerosol sources: biomass emissions contain large amounts of levoglucosan, plant sterols which vary by plant species, and other organic compounds [Simoneit, 2002]. Detailed organic species can also be used to differentiate between the different types of biomasses burned.
 A study of the emissions from biofuels burned in a traditional single-pot mud stove widely used for cooking food in South Asia produced aerosols with chemical properties different from previous studies of biomass. Habib et al.  demonstrated that biofuel burn rates (mass of fuel burned per hour, kg h−1) significantly influence aerosol composition and that these stoves produced aerosol containing more EC than OC. Due to the widespread use of biomass fuels in South Asia and the chemical nature of these emissions, biofuel use was consequently estimated to be the largest contributor to ambient EC in the region [Venkataraman et al., 2005]. In this study, emissions from these South Asian biofuels were characterized by molecular marker species to aid in the creation of biofuel source profiles for use with source apportionment studies of ambient aerosol in the region.
 A contemporary approach to the development of novel source profiles includes their immediate application and assessment. Such an approach was taken by Lee et al.  in the chemical characterization of particulate matter generated by open forest burning and by Lough et al.  in the development and application of molecular marker–based motor vehicle profiles. The combination of profile development with application demonstrates profile efficacy, documents caveats, and conveys their intended uses. Additionally, the comparison of new profiles to those documented in the literature further enhances the understanding of how biomass burning emissions vary across different regions and fuel types. The variability of biomass profiles and uncertainty of apportionment results can be evaluated using sensitivity analysis [Robinson et al., 2006; Sheesley et al., 2007], which compares apportionment results across solutions using different profiles.
 This study applies the newly developed South Asian biofuel profiles in a study of the sources of carbonaceous particulate matter (PM2.5) collected in Godavari, Nepal located in the foothills of the Himalayas on the southern edge of the Kathmandu Valley. The Godavari site was selected because it is representative of regional aerosol in South Asia where biofuel combustion is the dominant form of biomass burning according to emissions inventories [Streets et al., 2003b].
 The Himalayan region is subject to heavily light-absorbing aerosols that cause significant reductions in surface radiation, particularly during winter months [Ramana et al., 2004]. Himalayan radiative forcing is among the greatest in the world and may cause alternations of the region's hydrologic cycle [Ramanathan and Ramana, 2005]. A previous study of Himalayan aerosol composition was carried out by Shrestha et al.  at a remote site (4100 m altitude) and a rural Middle-Mountain site (1900 m altitude) and showed seasonal trends in water-soluble ion concentrations that were attributed to large-scale circulation patterns; aerosol concentrations were lowest during the summer monsoon, elevated during the dry winter, and highest during the premonsoon season. Aerosol species were highly intercorrelated and thought to be part of a thoroughly mixed air mass that may have traveled a long distance. A study during the premonsoon and monsoon seasons at the Pyramid Observatory in the Himalayas (5100 m altitude) in 2006 recorded fine particle concentrations elevated in comparison to the background atmospheric conditions and noted temporal consistency between elevated fine particle mass, EC, and ozone [Bonasoni et al., 2008].
 On a regional scale, Adhikary et al.  demonstrated that typical air masses impacting the Godavari site were westerly for dry months and southeasterly (from the Bay of Bengal) for the period of the monsoon. During periods of high organic and black carbon, air masses originated in the northwestern states of India. Several other studies of fine particle transport to the Kathmandu Valley have pointed toward long-range transport of pollution from Terai region [Hindman and Upadhyay, 2002; Shrestha et al., 2000], which includes the Himalayan foothills in southern Nepal and the Indian state of Uttar Pradesh. Local meteorology in the Kathmandu Valley was studied by Panday and Prinn , who documented daily ventilation of the basin during the daytime and a build-up of anthropogenic pollutants during the morning and night. Emissions from within the valley recirculated within the basin during the morning and impacted the surrounding mountains during times of ventilation.
2.1. Source Sampling
 The biofuel and biomass samples discussed in this paper were collected during the study of primary aerosols from biofuel combustion by Habib et al.  where sampling methods are described in detail. Biofuels were selected to represent important fuels in India and are summarized in Table 1. Fuels were burned in a single-pot mud stove and emissions were collected isokinetically using a dilution sampler with dilution ratio of 10–40 and chamber residence time of 200–300 s. Fuel woods, like acacia and mango woods were burned at low and high rates of approximately 1 kg and 2 kg of fuel per hour, respectively.
Table 1. Summary of Indian Biomasses Burn Conditions and Emission Factors Discussed in This Studya
Burn Rate (kg h−1)
PM2.5 Emission Factor (g kg−1)
Carbon Content (% PM2.5)
Fuels were burned in a traditional single mud pot stove, commonly used in India. Carbonaceous aerosol represents the sum of elemental and organic carbon. Values are summarized from Habib et al. .
low burn rate
high burn rate
low burn rate
high burn rate
Animal and Crop Wastes
Cow dung cake
2.2. Ambient Sampling
 Fine particulate matter (PM2.5) was collected daily at the Integrated Centre for Integrated Mountain Development (ICIMOD) Godavari Training and Demonstration Site (27.59°N, 85.31°E, elevation: 1600 m), located on the southern edge of the Kathmandu Valley in the foothills of the Himalayas. This rural/near-urban site is part of the regional network of Atmospheric Brown Cloud (ABC) observatories [Ramanathan et al., 2007]. Godavari is located within 20 km of Kathmandu, Bhaktapur, and Patan, the three largest cities in the Kathmandu Valley. Within the Valley is Nepal's greatest concentration of brick kilns, numbering approximately 125 kilns at the time of this study. A map of the study site in relation to the surrounding cities, brick kilns, and land uses is shown in Figure S1 in auxiliary material Text S1. Many aerosol sources were expected to contribute to ambient particulate matter at Godavari, that included urban emissions from nearby cities, transportation (diesel vehicles and motor bikes are common to the area), residential biofuel use (wood, dung, crop wastes), agricultural waste burning, and brick kilns (that burn coal, wood, and biomass briquettes made of crop wastes in unknown quantities).
 A PM2.5 sampler was used to collect ambient particulate matter. The sampler consisted of a sample inlet, a cyclone used to select particles less than 2.5 μm in aerodynamic diameter, and filter holders, which were all composed of Teflon-coated aluminum (URG Corporation, Chapel Hill, North Carolina). Airflow through the PM2.5 cyclone was 16 L per minute (lpm), which was initiated by a vacuum pump and was controlled by critical orifices. Airflow through the filter holders was measured by a calibrated Rotometer before and after sample collection. From 1 January 2006 to 24 November 2006, the sampling device was housed in a hand-built casing made of wood. An identical, commercially available, sampling device (URG-3000 ABC) was used from 25 November 2006 to 31 December 2006, which had the advantage of collecting both PM2.5 and PM10. The commercially available sampler operated at a flow of 64 lpm split between two PM2.5 inlets operating at 16 lpm each and a PM10 inlet operating at 32 lpm. The hand-built and commercially available samplers were operated concurrently for a period of 1 month to demonstrate consistency in collection, after which the hand-built version was discontinued from use.
 Particulate matter was collected on two types of substrates: quartz fiber filters (47 mm, Tissuquartz, Pall Life Sciences, East Hills, New York) and Teflon filters (47 mm, Teflo Membrane, 2.0 μm pore size, Pall Life Sciences). Throughout the study, filters were prepared in the laboratory prior to sample collection and were managed using clean-handling techniques [Stone et al., 2007]. Field blanks were collected every fifth day, daily samples began at 1000–1100 local time, and the sample duration was approximately 23.5 h. Samples were not collected from 28 August to 8 September and 27 September to 8 October, which corresponded to sampler cleaning and maintenance. Samples were shipped frozen in a cooler to the laboratory in Madison, Wisconsin, where all chemical measurements were made.
2.3. Chemical Analysis
 The mass of PM2.5 was measured gravimetrically using a high-precision scale (Mettler Toledo, Inc., Colombus, Ohio) [Stone et al., 2007], for which the uncertainty was estimated at 10%. The OC and EC content of PM was determined with a thermal-optical instrument (Sunset Laboratory, Forest Grove, Oregon) using the base case ACE-Asia method [Schauer et al., 2003]. The OC and EC measurements used in the biofuel source profiles were collected by the University of Wisconsin-Madison as part of the laboratory intercomparison presented by Habib et al.  in order to promote consistency with ambient aerosol measurements.
 Organic species were measured in aerosol collected in Nepal and in biofuel combustion samples following a standard extraction and quantification protocol [Stone et al., 2008]. Ambient measurements were made in composite samples of quartz fiber filters that consisted of equal filter portions from sampling days in a given month. Filters were spiked with isotopically labeled recovery standards that included the following compounds: pyrene-D10, benz(a)anthracene-D12, coronene-D12, cholestane-D4, eicosane-D42, tetracosane-D50, triacontane-D62, dotriacontane-D66, cholesterol-D6, and levoglucosan-13C6. Filters were extracted by soxhlets using 250 mL dichloromethane followed by 250 mL methanol. Extracts were reduced in volume using rotary evaporation under vacuum followed by evaporation under nitrogen. The extract was derivatized in two ways: first by methylation with diazomethane and then by silylation (with N,O-bis-(trimethylsilyl)trifluoroacetamide with 1% Trimethylchlorosilane). Derivatized extracts and external calibration standards were injected into a gas chromatograph mass spectrometer, GC(6890)-MS(5973) (Agilent Technologies, United States), for quantification of a wide range of organic compounds that included n-alkanes, hopanes, polyaromatic hydrocarbons (PAH), aromatic acids, levoglucosan, and sterols. The analytical uncertainty of organic species measurement was estimated to be twenty percent, based on the recovery of standard compounds and the standard deviation of field blank concentrations.
2.4. Source Apportionment and Sensitivity Analysis
 Primary source contributions to carbonaceous aerosol in Godavari were calculated using chemical mass balance (CMB) modeling software available from the United States Environmental Protection Agency (EPA-CMB version 8.2). This program solved the effective-variance, least squares solution to a linear equation of ambient measurements equated to primary sources and their relative contributions to OC, EC, and fitting species [Watson et al., 1984]. The model assumed that input species were stable in the atmosphere between source and receptor, which has been shown to be a reasonable assumption based on comparison of CMB results with colocated measurements and other source apportionment techniques [Docherty et al., 2008; Jaeckels et al., 2007].
 The organic species included in the model were the molecular markers observed at Godavari: nonacosane, triacontane, hentriacontane, dotriacontane, tritriacontane, 17-α(H)-21-β(H)-hopane, picene, levoglucosan, and campesterol. The source profiles used in this study were considered to be the best available at the time of the experiment and were drawn from the region of study when possible. Motor vehicle profiles were developed in the United States through a large and comprehensive study of diesel emissions from engines of various ages, weights, and driving cycles [Lough et al., 2007]. While the control technologies required for diesel engines vary by location, the composition of fuels and motor oils are relatively consistent and yield comparable emissions. The coal combustion profile was collected in China from a small boiler and was considered to be the most relevant to the study region [Zhang et al., 2008]. The vegetative detritus profile [Rogge et al., 1993] is the only one of its kind available and was not expected to change significantly across locations. Biomass profiles not generated in this study included fireplace oak combustion in California [Schauer et al., 2001], open pine forest burning in Georgia [Lee et al., 2005], burning of rice straw in China [Zhang et al., 2007], and modern-stove combustion of Bangladeshi cow dung and biomass briquettes [Sheesley et al., 2003]. The difference between carbonaceous aerosol attributed to primary sources and the total measured concentration is referred to as being from ‘other sources’ and is discussed in sections 3.3 and 3.4. A sensitivity test was used to investigate the impact of biomass source profiles on the apportionment of OC and EC in Godavari, similar to a previous CMB study [Sheesley et al., 2007], in which biomass profiles were systematically varied while molecular marker species and other source profiles were held constant.
3. Results and Discussion
3.1. Development of Source Profiles
 This study characterized South Asian biofuels burned under real-world conditions in a residential cooking stove commonly found in the region; fuels are summarized in Table 1. The bulk chemical, microphysical, and optical properties of the aerosol generated in source testing was the subject of a previous publication [Habib et al., 2008] and revealed an important distinction between burning high and low rates, in which the latter produced more EC than OC [Venkataraman et al., 2005]. These results were different from those of previous studies of biomass burning that indicated OC was the dominant form of carbon in biomass emissions. The chemical composition of emissions is relevant to the radiative properties of aerosols. EC is of particular interest because it is light-absorbing and plays a key role in atmospheric heating, surface cooling, and climate forcing [Ramanathan et al., 2007].
 The biofuel emissions contained varying concentrations of carbonaceous aerosol, which varied considerably with burn rate. PM2.5 generated by acacia burning at low rates comprised 9% EC and 27% OC, compared to high rates that produced emissions that were 15% EC and 38% OC. The PM generated by mango wood burning at low rates was 37% EC and 22% OC, compared to at high rates when emissions were 11% EC and 45% OC. On average, crop and cow dung burning produced PM2.5 that was 5% EC and 45% OC, with the exception of emissions from jute straw burning that consisted of 14% EC and 41% OC. The concentrations of EC normalized to OC (or EC to OC ratio) for fuels characterized in this biofuel study are shown in Figure 1a. As previously documented, these values demonstrate a large range across different biofuel emissions [Habib et al., 2008]. For mango wood, this ratio was 1.6 for the low rate compared to 0.2 for the high rate, whereas acacia wood at low and high rates were 0.3–0.4. Emissions from crop and animal wastes had generally lower EC concentrations relative to OC, ranging from 0.04 for rice straw, 0.13–0.15 for mustard and soybean stalks and cow dung, to 0.3 for jute stalk. The EC concentrations observed in this study are higher than those observed in previous studies of biomasses in a modern stove by Sheesley et al.  (0.01 for cow dung and 0.02 for rice straw burned in a modern stove) and Zhang et al.  (0.06 for rice straw burning in a Chinese stove). The variability of EC to OC ratios in these profiles demonstrates that biomass emissions vary significantly by burn conditions and that EC:OC ratios alone cannot be used to differentiate between combustion sources [Habib et al., 2008; Venkataraman et al., 2005].
 Biofuels were characterized by molecular marker species, which are compounds unique to aerosol source categories, present in the aerosol phase, and relatively stable in the atmosphere such that they can be used for source apportionment of ambient aerosol [Schauer et al., 1996]. The molecular marker species measured in biofuels emissions were levoglucosan, animal sterols (stigmastanol, coprostanol), plant sterols (stigmasterol, β-sitosterol, campesterol), and C21-C32n-alkanes. Semivolatile species (e.g., PAH) were not reported because of their high potential for sampling artifacts when low dilution ratios are used in source testing. Concentrations of molecular markers, OC, and EC in biofuels are summarized in Table S1 in auxiliary material Text S1.
 A key molecular marker is levoglucosan, which is a major component of biomass burning aerosol. Plant cellulose breaks down by one of two temperature dependant pathways; at temperatures greater than 300°C, pyrolysis occurs and levoglucosan is the major product [Simoneit, 1999]. Levoglucosan concentrations observed in this study ranged from 200 to 290 μg mgOC−1 for woods at high burn rates compared to 60–120 μg mgOC−1 for the same woods at low burn rates, as shown in Figure 1b. The observed trend is that the emissions rates for levoglucosan and OC decreased at low burn rates compared to high burn rates. The differences in levoglucosan concentrations at different burn rates were likely related to combustion temperatures, which ranged from 370 to 470 for low-rate burns and 630–670 for high-rate burns [Habib et al., 2008]. A similar shift in levoglucosan concentrations was documented in a previous study that compared emissions from fireplace to wood stoves under catalytic and noncatalytic modes [Fine et al., 2004a]. Levoglucosan concentrations ranged from 25 to 140 μg mgOC−1 for crop and animal wastes. Woody stalks from mustard and soybean crops had higher concentrations than rice straw or hollow jute stalks. Previous studies have reported similar levoglucosan concentrations in emissions from various types of biomass burning: 230–260 μg mgOC−1 in the emissions from fireplace wood burning [Schauer et al., 2001], 95 μgOC−1 in an open burn of an pine forest [Lee et al., 2005], 110–410 μgOC−1 from woodstove combustion of common American woods [Fine et al., 2004a], and 20–100 μgOC−1 in stove burning of Bangladeshi biomasses [Sheesley et al., 2003]. The results of this study and comparison to previous ones demonstrate a trend in which the burn type and conditions play a larger role in determining the bulk chemical composition, especially EC, and primary molecular marker concentrations, namely levoglucosan, than does the type of wood or crop waste burned.
 The presence or absence of molecular markers can provide information about specific fuel type in a broad category like biomass burning. Several animal sterols, stigmastanol and coprostanol were proposed by Sheesley et al.  to be unique markers of cow dung burning. This study confirmed previous results when cow dung markers were observed only in aerosol from this source, as shown in Figure 1c. Plant sterols, including stigmasterol, β-sitosterol, and campesterol, are shown in Figure 1d and are produced by vegetation in varying ratios, which can be useful in differentiating between types of biomasses [Simoneit, 2002]. Previous studies have shown campesterol to primarily be associated with grasses [Simoneit, 2002]; in this study, however, it was observed in every wood burning sample, albeit at lower concentrations than in rice straw, mustard stalks, and soybean stalks. The presence of plant sterols in cow dung emissions probably reflects the animal's diet [Sheesley et al., 2003] or the mixing of plant materials like straw during formation of dung patties. Biomass emissions also contain a homologous series of n-alkanes with an odd-carbon preference originating from plant waxes that can also be released to the atmosphere directly through abrasion of leaf surfaces as vegetative detritus [Simoneit, 2002]. The concentrations of n-alkanes shown in Figure 1e demonstrate variability with fuel type and were expectedly low in wood burning tests. Source profiles for biofuels were developed based on EC and molecular marker concentrations normalized to OC and are intended for use with source apportionment modeling in regions where biofuel combustion is prevalent.
3.2. Aerosol Composition and Seasonal Variation
 Monthly average PM2.5 mass, organic matter, and elemental carbon concentrations in Godavari, Nepal are shown in Figure 2. The yearly average PM2.5 concentration (±1 standard deviation), calculated from 24 h measurements (n = 326), was 26 ± 19 μg m−3. The 24 h average mass concentrations ranged from a maximum in mid-January of 120 ± 10 μg m−3 to a minimum in August of < 0.8 μg m−3. The annual average mass concentration in Godavari were similar to a nearby rural Himalayan site [Bonasoni et al., 2008], but were elevated with respect to pristine locations. Levels were on par with suburban sites in South Asia [Hopke et al., 2008] and were several times lower than in Indian megacities [Chowdhury et al., 2007]. On average, carbonaceous aerosol accounted for 40 ± 2% of PM2.5, when OC was converted to organic matter using a factor of 2.0 to account for the weight of other elements (oxygen, nitrogen, hydrogen, etc.) associated with organic carbon [Turpin and Lim, 2001]. The yearly average OC and EC concentrations (±1 standard deviation) were 4.8 ± 4.4 μgC m−3 and 1.0 ± 0.8 μg m−3, respectively. Over the course of the year, 24 h average OC concentrations ranged from 0.30 ± 0.15 to 33.0 ± 1.7 μgC m−3 while EC concentrations ranged from < 0.05 μg m−3 to 4.1 ± 0.7 μgC m−3. The remaining PM2.5 mass was expected to include dust and other inorganic constituents identified in Himalayan aerosol by Shrestha et al. , such as sulfate, nitrate, ammonium, potassium, and calcium.
 The seasonal trends of PM2.5 observed in Godavari were similar to previous studies in the Himalayas [Bonasoni et al., 2008; Shrestha et al., 2000], where mass concentrations were greatest in the dry winter months and were lowest during the summer monsoon. Carbonaceous aerosol concentrations peaked in late March and early April during the dry premonsoon season. Similar annual trends have been observed in urban areas in India [Chowdhury et al., 2007] and in Maldives during periods of transport of polluted air masses from South Asia [Stone et al., 2007]. The seasonal variation and aerosol loadings observed in Godavari were consistent with the annually occurring a regional haze in South Asia [Ramanathan et al., 2007]. A pollution episode occurred mid-June, well into the monsoon season, that caused levels of PM2.5 to exceed 40μg m−3 for 5 days. These pollution episodes in 2006 were also observed at the ABC Pyramid site in the higher Himalayas by Bonasoni et al. .
 Molecular markers for primary aerosol sources were quantified in composite samples and monthly concentrations are presented in Table 2 for the purpose of identifying and quantifying aerosol sources. The presence of elemental carbon and PAH were generic indicators that combustion of carbonaceous fuels contributed to ambient aerosol. As previously discussed, the presence of levoglucosan at concentrations 20–370 ng m−3 indicated the presence of biomass burning aerosol. Stigmastanol, an indicator of cow dung, was observed in every composite sample at concentrations of 0.3–0.9 ng m−3. A homologous series of n-alkanes with odd-carbon preference indicated the presence of modern biomass material, namely plant wax [Simoneit, 1999]. One or more hopanes in a series of three were observed in each sample from Godavari. Hopanes are associated with combustion of geologically aged fuels, which include coal combustion, fuel oil burning, and motor vehicles [Simoneit, 1999] while picene, a specific PAH with concentrations of 0.05–0.13 ng m−3 was a direct indicator of the presence of emissions from coal-burning [Oros and Simoneit, 2000]. The seasonal variability of molecular markers generally followed that of OC. Specifically, levoglucosan, sterols, n-alkanes, and hopanes were at peak levels in March and April, whereas PAH were at highest levels in January and February. Minimum molecular marker concentrations were observed in the summer months when aerosol loadings were lowest.
Table 2. Ambient Concentrations of Molecular Markers Observed at Godavari, Nepal, in 2006a
 Carboxylic acids were observed in relatively high concentrations in Godavari, although they are typically not considered molecular markers because they are not source specific. They have been associated with emissions from primary aerosol sources [Simoneit, 1985] and secondary sources via diurnal trends in urban areas [Fine et al., 2004b] and seasonal trends in remote locations [Sheesley et al., 2004]. In Godavari, wintertime concentrations of 1,2-benzene dicarboxylic acid ranged from 5 to 13 ng m−3 while 1,2,4-benzenetricarboxylic acid ranged from 0.72 to 2.0 ng m−3. Maximum carboxylic acid concentrations occurred in April at levels of 90 and 1.9 ng m−3 that were so high in comparison to other months that it was expected they were impacted by local point sources. Aromatic acid concentrations were markedly lower during the summer monsoon season and were not detectable in June. Aromatic carboxylic acids were noted to be in high concentrations in a polluted air mass affecting the South Asian region in the wintertime of 2004–2005 [Stone et al., 2007], which suggested that these compounds may be characteristic of regional air pollution in South Asia.
3.3. Base Case Source Apportionment of Organic Carbon
 Chemical mass balance (CMB) modeling was used to apportion fine particle OC and EC in Godavari, Nepal, to primary sources. The base case source apportionment results for OC are shown in Figure 3 and summarized in Table 3. EC apportionment is discussed in section 3.4. Primary sources included in the model were biomass burning, cow dung burning, coal combustion, vegetative detritus, and motor vehicles (i.e., diesel engines). The category “other” represented sources not included in the model, which could have been unidentified primary and/or secondary sources. The acacia low-rate profile was selected as the base case for biomass burning for several reasons. This tree species was commonly found in the tropical regions of Nepal and India, which were expected to have produced air masses that impacted Godavari. This profile had EC concentrations in between other high- and low-rate profiles and provided an intermediate solution representative of both high-EC emissions from residential stoves and high-OC emissions from open burning. Similarly, it had average levoglucosan concentrations that embodied a diverse range of burning conditions. The acacia low profile was found to be the most representative of residential biofuel use and the open burning of biomasses and agricultural wastes, which were all expected to be relevant sources in the region surrounding Godavari.
Table 3. Base Case Source Contributions Calculated by Molecular Marker Chemical Mass Balance Modeling to Ambient Organic Carbon at Godavari, Nepal, in 2006a
Diesel Engines (μgC m−3)
Coal Combustion (μgC m−3)
Vegetative Detritus (μgC m−3)
Cow Dung (μgC m−3)
Biomass Burning (μgC m−3)
Other (μgC m−3)
Values in bold are statistically significant.
0.32 ± 0.11
0.20 ± 0.04
0.32 ± 0.06
0.005 ± 0.006
1.71 ± 0.50
4.41 ± 0.73
0.72 ± 0.13
0.16 ± 0.04
0.36 ± 0.06
0.013 ± 0.006
1.20 ± 0.35
5.62 ± 0.67
0.28 ± 0.14
0.18 ± 0.04
0.38 ± 0.07
0.005 ± 0.010
2.85 ± 0.83
5.58 ± 1.04
0.35 ± 0.12
0.19 ± 0.04
0.29 ± 0.05
0.009 ± 0.007
1.82 ± 0.53
5.53 ± 0.78
0.33 ± 0.06
0.13 ± 0.03
0.15 ± 0.03
0.006 ± 0.004
0.80 ± 0.23
1.86 ± 0.40
0.15 ± 0.04
0.12 ± 0.02
0.15 ± 0.03
0.011 ± 0.004
0.37 ± 0.11
2.24 ± 0.33
0.06 ± 0.02
0.10 ± 0.02
0.03 ± 0.01
0.008 ± 0.002
0.15 ± 0.05
1.79 ± 0.27
0.06 ± 0.03
0.04 ± 0.01
0.05 ± 0.01
0.006 ± 0.002
0.33 ± 0.10
1.32 ± 0.27
0.03 ± 0.04
0.08 ± 0.02
0.09 ± 0.02
0.005 ± 0.003
0.73 ± 0.21
1.12 ± 0.34
0.05 ± 0.06
0.15 ± 0.03
0.21 ± 0.04
0.002 ± 0.003
0.90 ± 0.26
2.91 ± 0.45
0.08 ± 0.06
0.11 ± 0.02
0.16 ± 0.03
0.004 ± 0.004
0.90 ± 0.26
3.80 ± 0.49
0.07 ± 0.09
0.16 ± 0.03
0.24 ± 0.04
0.002 ± 0.006
1.77 ± 0.52
4.40 ± 0.75
 The base case CMB source apportionment results for Godavari in 2006 indicated that biomass combustion accounted for 21 ± 2% of OC (average ± standard error). Diesel engines made a smaller contribution to OC with 3.9 ± 0.8 and coal combustion contributed 3.0 ± 0.3%. Brick kilns, the only known industry to burn coal in the vicinity of Godavari, were expected to be the source of these emissions. Vegetative detritus accounted for 3.9 ± 0.3% of OC, while cow dung contributed less than 1%. Primary source contributions followed the same annual trends as their key molecular markers. The percent contribution of biomass burning to OC was lowest at the height of the summer monsoon (7–19%) and was greatest in March (31%) and September (35%) prior to and following the monsoon. Diesel engines and coal combustion had temporal trends in which percent contributions were greatest during the summer monsoon, which indicated that these anthropogenic sources did not vary seasonally like biomass burning.
 Other sources of OC, which correspond to the difference between modeled primary sources and total measured OC, were significant; they accounted for 54–84% of OC by month. Meanwhile, all of the measured EC was apportioned by the model. Other sources of aerosol, then, contained OC but not EC, and this suggested that other sources were not combustion-related. Previous studies of ambient aerosol have associated SOA with OC unapportioned to primary sources by the CMB model [Sheesley et al., 2004; Stone et al., 2008] and by related techniques [Engling et al., 2006]. The assumption that other sources are SOA was used to estimate SOA in Godavari and can be considered valid when all primary sources are identified and well characterized.
 The hypothesis that unapportioned OC is SOA was supported by the high levels of aromatic carboxylic acids previously discussed and consistency with previous studies in which SOA was considered to be an important contributor to PM2.5. Table 4 presents a summary of results from CMB source apportionment studies and aromatic carboxylic acid concentrations from Hong Kong [Zheng et al., 2006], an urban and peripheral site in Mexico City, Mexico [Stone et al., 2008], Michigan's Upper Peninsula in the United States [Sheesley et al., 2004], suburban Pensacola, Florida [Zheng et al., 2002], rural [Schauer et al., 2002] and urban [Docherty et al., 2008] locations in Southern California, and Maldives in the Northern Indian Ocean [Stone et al., 2007]. The comparison of the results from Godavari to these studies demonstrates that a large fraction of OC in nonurban areas is not apportioned to primary sources. Also, these data show that elevated concentrations of carboxylic acids generally increase along with other sources of OC, suggesting a common origin, presumably SOA. The seasonal trends in other sources at Godavari were similar to other Asian locations, specifically a rural site in Hong Kong where wintertime concentrations were greatest and summertime lowest. These seasonal trends were distinctly different from those observed in the United States where aromatic carboxylic acids and other sources were greatest in summer and autumn. The observed trends in Asia are partially governed by the summertime monsoon that corresponds to lowest aerosol concentrations and may also be related to the availability of photochemically active precursor and/or transport of pollution to Godavari.
Table 4. Summary of Ambient Aromatic Carboxylic Acid Concentrations and Organic Carbon Unapportioned to Primary Aerosol Sources From a Number of Source Apportionment Studies When Secondary Organic Aerosol was Considered to be an Important Source of Ambient Organic Carbona
 The stability of the source apportionment at Godavari was assessed by comparing the CMB source apportionment results when different biomass profiles were used. The examined profiles included six biomass profiles generated in this study and four profiles from the literature: fireplace combustion of oak [Schauer et al., 2001], open burning of a pine forest [Lee et al., 2005], stove combustion of rice straw [Zhang et al., 2007], and stove burning of biomass briquettes [Sheesley et al., 2003]. The biomass profiles generated in this study were limited to the base case (acacia burning at a low rate), acacia at a high rate, mango at a high rate, rice straw, mustard stalk, and soy stalk. Mango wood burning at a low rate was assessed separately and is discussed below. While all other variables were held constant, the biomass source profile was systematically varied and the source contributions to OC in Godavari using various biomass profiles are presented in Figures 4a–4b. The sensitivity of the model results were assessed for 2 months in contrasting seasons: January in the wintertime and August during the summer monsoon. Most estimates of biomass contributions to OC were clustered around the base case result which was 25% (1.7 ± 0.5 μg m−3) in January and 19% (0.33 ± 0.10 μg m−3) in August. The open burn profile yielded the highest estimates of biomass contributions to OC in both wintertime (53%) and summertime (36%). Low-end estimates were produced by profiles for acacia burning at a high rate, mustard stalk, and fireplace combustion of oak and ranged from 9 to 14% in the wintertime and from 7 to 10% in the summertime. When the four extreme cases were excluded, the average across the remaining cases was 25 ± 6% (±1 standard deviation) in the wintertime and 19 ± 4% in the summertime. These results indicated that biomass contributions to OC were self-consistent when various biomass profiles were used and most solutions were in strong agreement with the base case result. It was expected that the Godavari site was impacted by both biofuel burning and open burning in 2006 and that actual contribution fell in between these two extremes and within the projected uncertainty.
 The amount of OC apportioned to biomass burning sources directly impacted the OC represented by other sources. In the sensitivity analysis, the estimated contribution of other sources in January was 63% (4.4 ± 0.7 μg m−3) and the average across all other solutions was 65 ± 14%. In August, the base case contribution from other sources was 73% (1.3 ± 0.3 μg m−3) while the average across other solutions was 75 ± 9%. Other source contributions to OC were consistent across the different source apportionment solutions and clustered closely to the base case solution. As previously discussed, other sources were found to be consistent with SOA and the magnitude of its estimated contributions to OC in Godavari warrants further study of SOA precursors and formation mechanisms in South Asia.
 The selection of biomass profile had relatively small impacts on other sources. Diesel contributions to OC were 7 ± 1% in January and 5 ± 1% in August whereas coal combustion contributions were relatively stable at 2–3% in both months. Contributions from cow dung burning were consistently less than 1%, when either the cow dung profile generated in this study or that of Sheesley et al.  was used. The selection of biomass profile affected the estimation of contributions from vegetative detritus and cow dung, since all of these sources were associated with n-alkanes. The biomass profiles for open burning and crop wastes (rice straw, mustard stalk, and soy stalk) were associated with decreased contributions from vegetative detritus and cow dung as a result of this colinearity, also noted by Sheesley et al. .
 The apportionment of EC to biomass burning in Godavari exhibited a high degree of instability when different biomass profiles were used as shown in Figures 4c–4d. The base case EC contribution from biomass burning in January was 38% (0.53 ± 0.05 μg m−3) while the range of other solutions was 3–23% with an average and standard deviation of 15 ± 11%. Similar instability was observed for the month of August when the base case EC contribution from biomass was 37% (0.10 ± 0.01 μg m−3) and other solutions ranged from 3 to 22% with an average of 13 ± 11%. The base case profile represented the upper estimate of biomass contributions to EC, but was considered to be reasonable based on the moderate EC to OC ratio compared to other low burn rate profiles and prevalence of biofuel use in the region [Venkataraman et al., 2005]. Diesel engines accounted for the majority of EC not apportioned to biomass, contributing 81 ± 11% in January and 78 ± 11% in August. Coal combustion consistently contributed 3–4% of EC.
 The upper limit of biomass contributions to EC were explored using the biomass profile for mango wood burning at a low rate. This profile represented an extreme case with an EC to OC ratio of 1.6. The only other profile in the model with an EC to OC ratio greater than one was diesel engines with a ratio of 2.6 [Lough et al., 2007]. When profiles for low-rate mango wood burning and diesel engines were both included in the CMB model, the diesel engine contribution was forced to be negative. A statistically significant negative source contribution was considered physically impossible and rejected. In order to assess what fraction of biomass combustion could have been associated with biomass emissions high in EC relative to OC, weighted-average profiles of mango wood at low and high rates were assembled ranging from 0 to 40% low burn rate, by 10%. When the low burn rate emissions profile exceeded 40%, biomass contributions to OC and EC incrementally increased while diesel contributions were forced to be negative, providing unviable solutions. Source apportionment results for these profiles are shown in Figure 5. The maximum model-predicted EC contribution from biomass under these conditions was estimated to be 93% in the wintertime and 88% in the summertime, while minimum diesel engine contributions were 2% and 3%, respectively. From this, we confirm that the source apportionment of EC in Godavari is highly unstable due to the diversity of biomass burn conditions in the region. While the newly developed biofuel profiles contribute to a better estimate of biofuel contributions to ambient OC in South Asia, they could not improve the apportionment of EC. The biofuel emissions high in EC relative to OC may have accounted for a significant portion of ambient EC, as suggested by previous studies based on emissions inventories [Streets et al., 2003a; Venkataraman et al., 2005] and implied by the wide range of viable solutions in this study.
 Aerosol sources impacting Godavari in 2006 were diverse and expectedly came from a mixture of many source categories present within the South Asian region. CMB modeling provided an approach to quantitatively assess the current understanding of source contributions from biomass burning and fossil fuel combustion to carbonaceous aerosol (OC and EC) at the receptor location. Source profiles were selected to represent important sources in the region of interest, and included profiles for residential biofuel use and open biomass burning. The best estimate or base case result indicated that biomass burning and SOA were important sources of ambient OC whereas biomass burning and fossil fuel combustion were significant sources of EC. The systematic comparison of a range of possible solutions revealed that primary sources and SOA contributions to OC were clustered together among different model solutions and indicated a high stability in apportionment results. However, source contributions to EC were variable when different biomass profiles were used and revealed that source contributions to EC could not be reconciled with molecular marker CMB modeling in regions like South Asia, where biomass burning emissions high in EC relative to OC were expectedly important. To improve understanding of OC sources in remote Asian locations, studies should target SOA tracer compounds linked to biogenic and anthropogenic precursors [Kleindienst et al., 2007] and further characterization of water-soluble OC components. Further reconciliation of EC sources in South Asia should rely upon other apportionment techniques, such as multivariate analysis [Stone et al., 2007] and stable-isotope analysis [Gustafsson et al., 2009], which provide more constrained approaches to distinguishing between fossil and modern sources of EC.
 This research was supported through Atmospheric Brown Cloud (ABC) grants from the National Oceanic and Atmospheric Administration (NOAA). We thank the Wisconsin State Laboratory of Hygiene and scientists Jeff DeMinter and Brandon Shelton for assistance with chemical analysis and University of Wisconsin-Madison students Catherine Kolb and Marya Orf for technical support. We also thank Bhupesh Adhikary at the University of Iowa for helpful communications and the MENRIS division at ICIMOD for producing the map of the Kathmandu Valley. We appreciate the contributions of the Nepal Ministry of Environment and the United Nations Environmental Program Regional Resource Center for Asia and the Pacific for their support to the observatory program in Nepal.