The field office was located at the SOMAARTH Demographic, Development, and Environmental Surveillance Site in Palwal District, Haryana, India, run by the International Clinical Epidemiological Network (INCLEN). The site consists of 51 villages in the area with roughly 200 000 inhabitants (Balakrishnan et al., 2015; Mukhopadhyay et al., 2012; Pillarisetti et al., 2014).

Samples were collected from cookstove emissions between 5 August and 3 September 2015. Cooking events occurred at a village kitchen in Khatela, Palwal District. A local cook was hired to prepare meals for human consumption consisting of either chapati or rice with vegetables using a chulha stove, as well as animal food using an angithi stove.

Animal fodder simmers in a pot set upon smoldering dung in a clay bowl, referred to as an angithi. Chulha stoves are made from bricks and a covering of clay, and the availability of oxygen from the packing of biomass fuels results in primarily flaming combustion. The chulha is used to cook most meals for families in this village. Buffalo and cow dung patties and brushwood, in the form of branches and twigs, were used in chulha stoves, and for the 13 mixed fuel cooking events dung and brushwood were combined in a ratio determined by the cook's preference. Stoves and food ingredients were produced and fuels procured by the household or village. Fuel moisture content, fuel mass burned, and meals cooked were noted for each cook fire and can be found in the Supplement. Additional information regarding the cooking events and setup can be found in Fleming et al. (2018).

The sampling scheme is illustrated in Fig. 1. Air sampling pumps (PCXR-8, SKC Inc.) created a flow of emissions through the sampling apparatus. Emissions were captured with three-pronged probes that were fixed 60 cm above the cookstove. PM2.5 emissions and gases were sampled through cyclones (2.5 µm, URG Corporation). The resulting flow of particles was captured on either quartz or polytetrafluoroethylene (PTFE) filters, while gases were collected in an 80 L Kynar bag throughout the entire cooking event. Flows were measured both before and after sampling to ensure they did not change by more than 10 % using a mass flowmeter (TSI 4140). After the cooking event, pumps were turned off and a whole-air sampler, consisting of a stainless-steel canister (2 L) welded to a bellows-sealed valve (Swagelok), was filled to ambient pressure from the Kynar bag. Whole-air samplers were thoroughly flushed and evacuated in the Rowland–Blake laboratory before being shipped to India. At the end of the measurement campaign, whole-air samplers were shipped back to the laboratory and analyzed within 2 months of the end of the field measurements. A “grab” whole-air sample (WAS) was collected before cooking commenced each day. This served as a background for all cooking events sampled on that day.

Colman et al. (2001) described the VOC analysis protocol in detail. Briefly, a known amount of the WAS flowed over glass beads inside a U-shaped trap cooled by liquid nitrogen. The flow was regulated by a mass flow controller and resulted in a known, roughly 600 Torr drop in the pressure in the whole-air sampler. High-volatility gases such as O2 and N2 passed over the beads, while lower-volatility gases adsorbed onto the beads. Compounds were re-volatilized by immersing the trap in hot water and were injected into a He carrier gas stream where the flow was split equally to five columns housed in three gas chromatographs (HP-6890). The compounds were separated by gas chromatography (GC) and subsequently detected by two electron capture detectors, two flame ionization detectors, and one quadrupole mass spectrometer detector. Peaks corresponding to compounds of interest were integrated manually. CO, CO2, and CH4 were analyzed using two GC systems equipped with thermal conductivity and flame ionization detectors as described in Simpson et al. (2014). The CO and CO2 GC with flame ionization detection (GC-FID) system is equipped with a Ni catalyst that converts CO into detectable CH4.

Emission factors (EFs) were calculated using the carbon-balance method, which assumes that all carbon in the fuel is converted to CO2, CO, VOCs, and PM when the fuel is burned. The total gas-phase carbon emissions were approximated with the concentrations of CO2, CO, and 76 detected VOCs measured using WAS. This is a good approximation since most emitted carbon resides in CO2 and CO (>95 %), so the error associated with VOCs that are not detected is relatively small (Roden and Bond, 2006; Smith et al., 1993; Wathore et al., 2017; Zhang et al., 2000). The mass of carbon in species i (mi,C) was calculated using Eq. (1): mi,C(g)=Ci,C(gm-3)CCO2,C+CCO,C+CCH4,C+…+CC6H6,C(gm-3)⋅mT,C(kg)⋅1000g1kg, where Ci,C represents the mass concentration of carbon for species i, and mT,C refers to the carbon mass of the fuel, adjusted for ash and char carbon. Fuels were weighed before they were burned, and the dry mass was calculated based on moisture content measurements. The ash was weighed after the cooking event and subtracted from the dry mass of the fuel giving the net dry fuel burned for the cooking event, mT. When mixtures of dung and brushwood were used, both were individually weighed to more accurately determine the carbon mass burned. The fraction of carbon in the fuel used to yield mT,C was taken to be 0.33 for buffalo and cow dung and 0.45 for brushwood fuels based on Smith et al. (2000a, b). Carbon in ash was estimated as 2.9 % and 80.9 % of the measured char mass for dry dung and dry brushwood, respectively (Smith et al., 2000b). For each species with nc,i carbon atoms in the formula and molecular weight (MWi), the emissions factor (EFi) was calculated using Eq. (2). EFigVOCikgfuel=mi,C(g)⋅MWi(gmol-1)nc,i×12.00(gmol-1)mT(kg) In addition to the emission factor normalized by the total fuel mass, emission factors were normalized to the total carbon mass in the fuel, calculated via Eq. (3). EFigVOCikgfuelC=EFigVOCikgfuel⋅mT(kg)mT,C(kg) PM2.5 mass was determined gravimetrically using Teflon filters (PTFE, SKC Inc., 47 mm) weighed on a Cahn-28 electrobalance with a repeatability of ±1.0 µg after equilibrating for a minimum of 24 h in a humidity- and temperature-controlled environment both before and after sample collection (average temperature: 19.5±0.5 ∘C; average relative humidity: 49±5 %). Another gravimetric filter was collected in the background during the cooking event and was equilibrated and weighed in the same way (Fig. 1, Teflon C). Five field blanks filters were prepared by opening filters and then immediately closing and sealing the filters the same way as all samples; these filters had negligible mass loading (average: 0.40 µg) relative to samples (average: 1.57 mg). The method detection limit was determined to be 9.3 µg from the standard deviation of the field blank filters multiplied by 3. The background filter mass (Fig. 1, Teflon C) was adjusted by matching the sampling volume to that of the sample gravimetric Teflon filter (Fig. 1, Teflon A). This assumes that the mass of PM2.5 collected on the filter is directly proportional to the flow rate through the filter. The background mass was then subtracted from the sample mass to obtain the mass of PM (mPM) in Eq. (4) below. EFPMEFCO=mPM/VairmCO/Vair

Modified combustion efficiency is defined as follows: MCE=ΔCO2ΔCO+ΔCO2, where ΔCO and ΔCO2 are background-subtracted mixing ratios of CO and CO2 for the time-integrated (WAS). The total carbon mixing ratio is approximated by the sum of carbon monoxide and carbon dioxide in this definition.

Relative SOA-forming potential from measured VOCs was estimated using secondary organic aerosol potential (SOAP) values from Derwent et al. (2010), who used a photochemical transport model to simulate the SOA mass increase from the instantaneous emission of a particular VOC in a single parcel of air traveling across Europe. The model was outfitted with the Master Chemical Mechanism (MCM v.3.1) and the UK National Atmospheric Emission Inventory. The model was initialized with 2 ppbv NO and 6 ppbv NO2 (Derwent et al., 1998). SOA mass was estimated assuming equilibrium partitioning of oxidation products. Partitioning coefficients were calculated using absorptive partitioning theory of Pankow (1994). In the SOAP approach, all SOA mass increases from a particular VOC (i) are normalized to that of toluene as shown in Eq. (6). SOAPi=incrementinSOAmassconcentrationwithspeciesiincrementinSOAmassconcentrationwithtoluene×100 SOA-forming potential was calculated from the published SOAP values using Eq. (7). SOApotential=∑i=0nSOAPi×EFigVOCikgfuelC Table S1 in the Supplement lists the SOAP values used to calculate SOA-forming potential in this study. We emphasize that the SOA-forming potential presented here is a relative value and does not represent an absolute SOA yield.

Total OH reactivity normalized by the mixing ratio of CO was calculated using Eq. (8). OHreactivity1sppbvCO=∑i=0nkOH,icm3molecs×ERipptvVOCippbvCO×2.46×107moleccm3pptv Second-order rate constants (kOH) at 25 ∘C were taken from the NIST chemical kinetics database (Manion et al., 2015). Table S1 reproduces the kOH constants used in the study. ERi is the emission ratio for compound i in pptv of VOC per ppbv of CO. The last term serves as a conversion factor from VOC mixing ratio to concentration at standard ambient temperature and pressure (25 ∘C, 1 atm). By using the emission ratio to CO, we can track OH reactivity from VOCs depending on the extent of dilution from the plume. From here forward, the OH reactivity (per second) reported is the average at the location of the sampling probes, or roughly 60 cm above the cookstove.

The ozone-forming potential was estimated from the incremental reactivity of VOCs tabulated in Carter (2010). Incremental reactivities were calculated by comparing ozone formation before and after a VOC was introduced in a box model simulation. The maximum incremental reactivity (MIR) scenario sets high NOx concentrations, optimized to yield the largest incremental ozone production. In other words, ozone production was VOC limited. Because of this, the OFPs given here represent a scenario where VOC emissions from cooking have the largest impact on ozone production. This high-NOx scenario was chosen for high sensitivity in ozone production from cooking emission VOCs. In addition, it is expected to be more realistic for a smoke plume in which NOx is co-emitted with VOCs. OFPs were calculated using Eq. (9). OFPgO3kgfuelC=∑i=0nMIRgO3gVOCi×EFigVOCikgfuelC OFPs used in this study are listed next to the corresponding compound in Table S1.

One-way analysis of variance (ANOVA) with Tukey post hoc testing was utilized to determine if there were significant differences (p<0.05) in emissions of specific VOCs among categorical variables, i.e., stove and fuel types. All analyses were performed in R version 3.4.0 and RStudio version 1.0.143 (RStudio Team, 2016).

Average VOC and PM2.5 EFs (g kg-1 dry fuel) as well as MCE are given in Table 1. The compounds are grouped by fuel–stove combination, with major species (CO2, CO, CH4, and PM2.5) listed first, followed by sulfur-containing compounds, halogen-containing compounds, organonitrates, alkanes, alkenes, alkynes, aromatics, terpenes, and oxygenated compounds. The sample size (n) used for calculating the average values and standard deviations was n=18 for dung–chulha, n=14 for brushwood–chulha, n=13 for mixed fuel–chulha, and n=10 for dung–angithi. For the majority of the compounds, the standard deviations are smaller than or comparable to the average values, indicating fair reproducibility. There are many factors that may lead to variability in biomass burning emissions, including pyrolysis temperature (Chen and Bond, 2010), fuel moisture content (Tihay-Felicelli et al., 2017), and the wind speed/direction (Surawski et al., 2015), among others. Relationships between emissions and fuel moisture content (Fig. S1) or meal cooked were not found to be significant for any compounds (all p<0.05). This paper therefore focuses on the relationships between emissions and fuel–stove combination.

Figure 2a visually shows the mass fraction attributed to each compound class for the measured gas-phase emissions. The total EFs given below the pie charts are normalized by fuel carbon in Fig. 2a in order to compare between cook fires generated with dung, wood, and wood–dung mixtures, which have different carbon contents. The total measured VOC emissions from dung–angithi were roughly twice those of dung–chulha in terms of gram per kilogram fuel carbon. Further, dung–chulha emitted more than twice the amount emitted by brushwood–chulha. The most prominent difference is non-furan oxygenates, making up almost half of all brushwood–chulha emissions and a smaller fraction for other fuel–stove combinations. While oxygenates make up a higher fraction of brushwood–chulha emissions, the absolute EFs for oxygenates from dung-burning and angithi cook fires are higher as discussed later in more detail.

Table 2 shows EFs (g kg-1 fuel C) for select VOCs. The differences in mean EFs for each fuel–stove combination are also included in Table 2. Mean differences in EFs reported for chulha and angithi stoves were calculated for cook fires utilizing only dung fuels. Likewise, mean EFs for wood and dung cook fires only represent cooking events using the chulha. This was done to isolate a single variable – either fuel or stove type. For all alkanes and most alkenes, we measured higher emissions for dung–angithi cook fires (Table 2). Also, from the mean differences in EFs, we found that stove-specific combustion conditions impact emissions more than the selection of fuel type. The difference is so dramatic for alkanes and most alkenes that the mean difference in EFs for cookstoves burning dung is always larger than the mean EF of that compound. For comparison, the mean difference in EFs for chulha cookstoves is always lower than the overall mean EF. Ethene was an exception; there was no relationship between ethene emissions and stove type. On the other hand, the mean EF of ethene by dung cook fires was very large compared to mean EFs from brushwood cook fires, with a mean difference in EFs of 4.05 g kg-1 fuel C. Some alkenes with two double bonds were also exceptions. For 1,3-butadiene (p=0.06) and 1,2-butadiene (p=0.089), stove and EF may or may not have a significant relationship. 1,2-Propadiene emissions from chulha cookstoves are higher (p<0.01). All three compounds still show a significant relationship to fuel type, with EFs being higher for dung cook fires.

Similar to alkanes and alkenes, aromatics, oxygenates, and halogen- and sulfur-containing compounds all had higher emissions per kilogram of fuel carbon when dung fuels and angithi stoves were utilized compared to brushwood fuels and chulha stoves, respectively. We focus on the behavior of the most interesting groups of compounds in the discussion below.

The chlorine-containing organic compounds are generally not expected to come from cook fires in large quantities. However, we observed an interesting practice in which the cook often used plastic bags to start the fire, which could be a source of chlorine-containing compounds if composed of polyvinyl chloride. Carbonyl sulfide (OCS) is largely responsible for the yellow sulfur-containing fraction in Fig. 2a, and biomass burning is a well-known source of OCS in the atmosphere (Crutzen et al., 1979). Similar to other VOCs, OCS was significantly emitted in higher quantities when angithi stoves and dung fuels were utilized.

Benzene had higher emissions from chulha stoves, which had higher MCEs when cooking with dung fuels compared to angithi stoves (dung–chulha: 3.18 g kg-1 fuel C; dung–angithi: 2.38 g kg-1 fuel C). As the simplest aromatic compound, benzene also had the largest average difference in fuel type EFs compared to other aromatics (2.18 g kg-1 fuel C, dung–wood). This information is relevant for exposure assessment, as benzene is a known human carcinogen. Higher benzene emissions from chulha cook fires could lead to higher benzene exposures, which is a potential public health concern. However, it should be noted that the cook usually cannot control the stove used, as the angithi and chulha are used to prepare different types of meals, and exposure to benzene is not straightforward from its emission factors.

Higher emissions of alkynes were observed from dung fuels and chulha cookstoves. The latter observation is consistent with the literature showing flaming combustion generates more alkynes (Barrefors and Petersson, 1995; Lee et al., 2005). Chulha cook fires always had higher MCE than angithi cook fires (Table 1), which rely on smoldering combustion. Approximately the same difference in alkyne emissions results from comparing the chulha to the angithi using dung, in relation to using wood versus dung in combination with the chulha. There were two exceptions in stove type for 1-butane (p=0.055) and 2-butane (p≫0.05). The former may or may not have a relationship with stove type, while the latter does not. Emissions of some compounds did not show a relationship with either fuel or stove type; they are listed in Table S2.

VOC emissions from Stockwell et al. (2016) are also provided in Table 1 for comparison of VOC EFs. Samples in Stockwell et al. (2016) were collected in April 2015 in and around Kathmandu and the Tarai plains, which border India. While both are EFs from cookstoves using similar fuels, there are differences in the studies that should be noted. Stockwell et al. (2016) collected measurements of simulated cooking in a laboratory and from cooking fires in households; it was not noted in the latter case what meals were cooked. EFs were calculated from similar WAS measurements, but as grab samples in an area of the kitchen away from the fire (as opposed to the time-integrated approach used here). Emissions were assumed to be well mixed in the kitchen prior to sampling. Stockwell et al. (2016) also used a range of stoves, including the traditional single-pot mud stove; open three-stone fire; bhuse chulo; and rocket, chimney, and forced draft stoves. “Dung” cook fires sometimes used a combination of fuels, such as wood. Finally, our study also has a larger sample size than Stockwell et al. (2016), with n=49 versus n≈10.

The emission factors for most compounds determined in this study were lower compared to those reported in Table 4 of the Stockwell et al. (2016) paper. Figure S2.1 visually shows that the EFs were generally lower in the present study. In some cases, EFs in this study were an order of magnitude lower, most notably n-pentane and n-hexane. We also found that our EFs were always higher for dung–chulha compared to brushwood–chulha, which was not always the case in Stockwell et al. (2016). The EFs in Stockwell et al. (2016) could be biased high due to calculations rather than real differences in emissions. For example, ignoring ash and char carbon and using the same carbon content inflates the EFs reported in our paper by 7 % for dung and 24 % for brushwood emissions. However, this is a small percentage compared to the observed differences between EFs between the two measurements. By examining the EFs reported in the supporting information section of Stockwell et al. (2016), we found that the disagreement resulted from the way the final recommended EFs were obtained from the measurements. Because Stockwell et al. (2016) measured both laboratory and field cook fires, they elected to adjust the laboratory EFs to account for the lower MCEs observed in the field. It has been shown by Roden et al. (2009) and Johnson et al. (2008) that cooking activities can strongly influence emissions, for example due to the cook tending to the cook fire differently and thus affecting combustion conditions. However, such adjustments should be done with caution because EF and MCE do not always follow a linear trend, as explained in the next section. Figure S2.2 plots the unadjusted laboratory and field emission factors as a function of MCE for all the measurements in Stockwell et al. (2016), as well as this study. Plotting the unadjusted EFs resolves most of the differences in observed EFs between the two studies. The data show an encouraging level agreement despite the differences in the experimental design between the two studies.

The use of dung and angithi, rather than brushwood and chulha, respectively, results in lower modified combustion efficiencies as shown in Fig. 3. In general, at lower MCEs we measured higher emissions of gas-phase compounds as discussed in Sect. 3.1. For example, emissions of ethane (Fig. 3a) and other alkanes increase with decreasing MCE. However the dependence of ethane EF on MCE is not linear as observed in previous studies (Liu et al., 2017; Selimovic et al., 2018). For other VOCs, the dependence of the EF on MCE deviates from the linear trend even more, with the maximum EF observed at intermediate MCE values. For example, the ethene EF (Fig. 3b) increases with decreasing MCE at MCEs >0.85, but it has the opposite trend at MCEs <0.85. Previously, we discussed that there is no relationship between ethene EF and stove type, and we see this more clearly in Fig. 3b. Alkynes have the same relationship to MCE as ethene, but it is even more pronounced (Fig. 3c). Benzene (Fig. 3e) stands apart from other aromatics with a relationship with MCE similar to ethene, while other aromatics have an EF-versus-MCE curve similar to alkanes and most other VOCs. In Fig. 3e, we see again that emissions from brushwood–chulha and dung–angithi cook fires result in lower emissions of benzene compared to dung–chulha. Alkenes with two double bonds generally have a negative correlation between emissions and MCE, such as 1,3-butadiene in Fig. 3f. The 1,3-butadiene EF-versus-MCE plot is not necessarily representative of all analogous plots for alkenes with two double bonds, as they have different shapes. 1,3-Butadiene was chosen as its emission is high compared to other compounds in its subcategory, and it also has health implications. It also happens to have a more linear relationship with MCE, albeit noisy.

It is of interest to compare EFs obtained from different fuel–stove combinations but with the same MCE. In the case of ethane, different cook fire types yield vastly different EFs at the same MCE. For example, at MCE ≈0.87, mixed fuel–chulha has an EF of roughly 1.5 g kg-1 fuel C, dung–chulha is 2.5 g kg-1 fuel C, and dung–angithi is 5.5 g kg-1 fuel C. Knowledge of the cook fire MCE alone is not sufficient to determine the EF of ethane. Combustion conditions specific to the fuel–stove combination are a significant factor in determining cook fire emissions. A similar conclusion can be reached for most of the measured gases, including non-ethene alkenes in Fig. 3d.