Emissions and Chemical Components of PM2.5 from Simulated Cooking Conditions Using Traditional Cookstoves and Fuels under a Dilution Tunnel System

Despite the considerable cost associated with estimating household emissions from solid fuel, which are frequently undetected by air quality monitoring systems, compiling such an inventory is critical to identifying the link between indoor pollution and health effects. Therefore, this study used the UP Diliman dilution tunnel system (UPDDTS) to characterize the composition of particulate matter in the smoke and quantify the PM2.5 emitted by traditional Philippine cooking systems, viz., a charcoal-burning cement stove (CCP), a sawdust-burning tin-can stove (KKP), a fuelwood-burning metal-grill stove (MFP), a kerosene-burning metal stove (MKP), and a charcoalburning metal-grill stove (MCC). Forty-three sampling tests revealed that water-soluble K+ (23.0 ± 1.9 μg m–3), Cl– (12.3 ± 1.0 μg m–3), and Na+ (43 ± 22 μg m–3) contributed to the majority of the ionic mass concentrations generated by the CCP and MKP, respectively, whereas levoglucosan— a signature of biomass burning—dominated the PM2.5-bound monosugars emitted by the KKP (78.72 ± 6.96 μg m–3), MFP (0.76 ± 0.34 μg m–3), and MCC (10.21 ± 2.64 μg m–3). The abundance of the water-soluble organic carbon (WSOC) in all of the samples, except those from the MKP, depended on the surface area—and thus the facet—of the fuel. Additionally, the elemental compositions of the PM2.5 from the CCP, KKP, and MCC mainly consisted of Pb (1.96 ± 1.04 to 76.02 ± 151.42 ng min–1), but those for the MFP and KKP primarily contained Cu (2.23 ± 1.18 ng min–1) and As (5.51 ± 1.08 ng min–1), respectively. The PM2.5 emission rates exceeded the World Health Organization (WHO)’s emission rate target guideline for ventilated conditions (0.8 mg min–1) by 1.9 × 106 to 23 × 106 mg min–1, and the highest PM2.5 emission factor, 0.032 ± 0.016 kg-PM2.5 kg-fuel–1 y–1, which was exhibited by the MKP, surpassed values in the literature by three orders of magnitude.


34
Household energy use remains to be directly linked to health risks, since people have always 35 relied on cookstoves and fuels of varying quality for daily supply of food, while smoke emitted 36 from the use of these cookstoves and fuels can compromise indoor air quality. According to the 37 United Nations Development Programme (UNDP), there are around 3 billion people that still rely 38 on wood, coal, charcoal, or animal waste for cooking and space heating (United Nations, 2018). 39 This translates to about 4.3 million premature deaths in 2012 globally due to household air pollution 40 from using solid fuels for household energy, with the highest percentage in the low-and middle-41 income countries (LMICs) (United Nations, 2018). In the Philippines, traditional cookstoves such 42 as cement stoves and metal grills, both of which use either charcoal or fuelwood, are still ubiquitous. 43 These traditional cookstoves are used for tenderizing meat, low-cost cooking (usually by the 44 underprivileged sector who cannot afford cleaner cookstoves), small-scale income generation, 45 meat-grilling, or simply cooking daily household food. Small-scale smoke houses are rampant in 46

A C C E P T E D M A N U S C R I P T
3 population in the Philippines are attributable to these indoor air pollutants (WHO, 2016;Yee, 2018). 53 Moreover, it was found to disproportionately affect the low-and middle-income households (Gloor, 54 2014). Pertinent diseases include the incidence of acute lower respiratory infections (ALRIs) and 55 chronic obstructive pulmonary disease (COPD). A few studies that focused on household exposures 56 from cooking in low-income countries have been conducted in the past, such as one by Saksena et 57 al. (2007) where 120 houses were subjected to a socioeconomic survey and 30 houses were sampled 58 for carbon monoxide and particulate matter (PM). With an estimate of over half of Filipinos being 59 exposed to these household indoor cooking pollutants and a relative lack of information on their 60 long-term exposure effects, the challenge of household air pollution issues remains to be tackled. 61 Furthermore, there is a need to create such awareness for the general public and policy makers. 62 In order to estimate emissions from household energy use, measurements may be performed 63

A C C E P T E D M A N U S C R I P T
8 the vessel inside the digester, and 20 minutes of cooling down the vessel outside the digester. The 143 maximum energy input was 1000 W. 144 A multi-elemental ICP-MS calibration standard (10 μg mL -1 in 5% HNO3 + tr HF, Peak 145 Performance Certified Reference Materials) was used for external calibration of 14 elements (Al,146 As, Ca, Co, Cd, Cr, Cu, Mg, Mn, Na, Ni, Pb, Sr, and Zn). Standard concentrations used were 0, 147 0.01, 0.05, 0.1, 0.5, 1, 5, 10, 50, 100, and 500 ppb. Calibration curves were generated using the 148 ICP-MS software and concentrations of the elements in the sample digestates were calculated in 149 this program. Individual blank PTFE filters were spiked with known amounts of the elements (0.1, 150 0.2, 0.25, 0.5, 0.75, and 1 mL of 10 ppm multi-elemental standard) and digested with the same 151 method used for the samples to conduct recovery tests. 152 153

Analysis of water-soluble ions, organic carbon, and cellulose degradation products 154
Ultrapure water used in this study was prepared using a Labpure S1 filter with a UV lamp, 155 with resistivity and total organic carbon (TOC) values of 18.2 MΩ cm -1 and 1 ppb, respectively 156 (PureLab Ultra, ELGA). A quarter of the filter was extracted with 12 mL organic-free ultrapure 157 water under ultrasonication for 30 minutes. Extracted samples were then filtered through a 0.45 158 µm syringe filter (Pall Gelman Acrodisc® ) to remove water-insoluble suspended materials.

A C C E P T E D M A N U S C R I P T
9 the water extract was used in the analysis of cations and anions using ion chromatography (Dionex 161 ICS-5000, Thermo Fisher Scientific respectively. Cation and anion standard concentrations used in the analysis were 0.2, 0.5, 1.0, and 167 2.0 ppm. Standards were measured before and after each analytical sequence of analysis. Eight (8) 168 spiked samples were also analyzed prior to test samples to validate method and instrument 169 performance. A spiked sample was also analyzed for every batch of 10 samples. 170 To measure water-soluble organic carbon (WSOC), a quarter of the filter was extracted with 171 20 mL ultrapure water in a glass vial using an ultrasonic extractor for 30 minutes. The water extracts 172 were filtered with a syringe filter (0.4 μm PTFE membrane, Pall Corp.) and then introduced to a 173 total organic carbon (TOC) analyzer (TOC-LCSH/CSN, Shimadzu). In the TOC analyzer, the water 174 extract was acidified using HCl and then bubbled with pure N 2 gas to eliminate the inorganic carbon 175 component. Then, the organic compounds in the water extracts were combusted at 680°C with a 176 platinum catalyst to form CO2, which is then quantified using a nondispersive infrared (NDIR)

A C C E P T E D M A N U S C R I P T
10 sensor. The instrument was calibrated with known amounts of potassium hydrogen phthalates 178 (C6H4(COOK)COOH) solution. where ρ the density of dry air at 1.0 atm and 20 °C (kg m -3 ), u is the velocity (m s -1 ), L is the

A C C E P T E D M A N U S C R I P T
11 hydraulic diameter (m), and is the viscosity of dry air (kg m -1 s -1 ) (Engineering ToolBox, 2003). 196 The air exchange rate per hour under the UPDDTS test section was calculated using the 197 general formula for air change per hour (ACH): 198 where Q is the volumetric flow rate (m 3 s -1 ) of the air flow inside the UPDDTS (in this case, the 200 average velocity, m s -1 , multiplied by the UPDDTS cross-sectional area, m 2 ), and 60 is a conversion 201 factor from seconds to hours. The ACH is then used to calculate the PM2.5 emission rates of each 202 stove and fuel combination tested. where Cs is the measured concentration of the spiked sample, C is the measured concentration of 216 the unspiked sample (background concentration), and S is the theoretical concentration of the 217 spiked sample. Method detection limit (MDL) was calculated using Eq. (5): 218 where t is the student's t-value at 99% confidence level (t7 = 3.14), and S is the standard deviation 220 of the seven spiked replicates.  Table 2, satisfying turbulent flow within the tunnel system. M A N U S C R I P T 14 In the UPDDTS, the distance between the test chamber and the sampling chamber is 2 248 meters (79′′ or ~8 duct diameters), at a flowrate of 2,700 m 3 h -1 and Re that ranged from 0.8 × 10 5 249 to 1.2 × 10 5 . This performance is an order of magnitude greater than Lawrence Berkeley National 250 Laboratory's dilution tunnel setup at a flow rate of 340 m 3 h -1 and Re at 3.9 × 10 4 (Wilson et al., 251 2017). Although the UPDDTS was not tested for accuracy using pulsed tracer gases, Wilson et al. 252 (2017) suggested that a dilution tunnel between 7-10 duct diameters satisfies turbulent flow and 253 would have well-mixed sampled emissions. However, a turbulent flow will also introduce particle 254 loss to the walls, which was not quantified in this study. This would mean that values presented 255 herein could possibly be lower than actual emissions in an open environment. However, one would 256 expect that losses may not be very significant when the tunnel is at a higher temperature such that 257 no condensation occurs on the walls of the tunnel. Hildemann (1989) reported that for a turbulent 258 dilution tunnel (Re = 10,000) with a large diameter, particle loss is only about 1-2%. 259 260

Particulate emission rates and emissions 261
The extent of impact to household air quality of emissions from cookstoves and fuels are 262 determined by the emission rates. Emission rate targets are established by correlating specific 263 health risks to emissions, and compliance to which may assess how well various interventions can 264 meet the air quality concentrations specified in the WHO guidelines. The WHO provided a PM2.5

A C C E P T E D M A N U S C R I P T
15 emission rate target (ERT) of 0.8 mg min -1 for cookstoves used under vented kitchen conditions. 266 If complied with, household air quality would meet the annual final PM2.5 guideline value of 10 μg 267 m -3 (WHO, 2014). When tested under the performance conditions of the UPDDTS, cookstoves and 268 fuels in this study range from 1.9 × 10 6 to 23 × 10 6 mg min -1 (Table 3), several orders of magnitude 269 higher than the WHO ERT at 0.8 mg min -1 . Hence, the design of the UPDDTS does not only 270 provide satisfying performance of the mixing of sampled air, but it also can provide a means to 271 determine whether the cooking equipment and fuel can meet the annual air quality guideline value 272 of 10 μg m -3 under a vented condition. 273 Emission factors for PM2.5 calculated using Eq. (7) are shown in Table 3 (Table 3), likely due to a smaller PM2.5 contribution (less than 1%) to the

A C C E P T E D M A N U S C R I P T
16 For the CCP and MFP, the calculated PM2.5 emission factors are in the same order of 284 magnitude and are therefore in reasonable agreement with the EMEP/EEA values. MFP is higher 285 by a factor of two, but CCP is lower than the literature value. The difference in the values in this 286 work is due to the difference in the type of organic matter or wood sampled and thus a 287 straightforward comparison may not be appropriate. 288 For MKP, the order of magnitude obtained using the UPDDTS is much higher than 289 literature but within the same order of magnitude calculated for the other sources. This is due to 290 the difference in the composition or grades of kerosene produced in the Philippines than those in 291 developed countries. Cleaner technologies allow for cleaner fuels to be produced in developed 292 countries; whereas crude burning of kerosene with simple, low-technology cookstoves allow for 293 emission of more PM, especially PM2.5, and thus make the user more vulnerable to pollutant 294 exposure. This shows that if literature values were used for emission inventories, it may cause 295 underestimations of emissions from kerosene. 296 Comparison of tested emission sources show that while KKP has the highest emission rate 297 per minute, MKP has the highest emission factor on a per-amount of fuel used in a year basis. The 298 high emission rate of KKP as compared to the other sources is caused introducing very fine, grainy 299 wood shavings, as opposed to fuelwood and charcoal having greater surface areas and therefore 300 would take more time to burn and give off smaller particulates. Thus, the size of the raw material

A C C E P T E D M A N U S C R I P T
(apart from the weight) can be considered as one of the factors that affect the amount of particulate 302 matter emitted by a certain emission source. However, as for kerosene, the amount of particulate 303 matter emitted is dependent on its density. It should be noted, however, that the filter for kerosene 304 was the blackest, which shows more black carbon (incomplete combustion) as kerosene is purely 305 hydrocarbon. However, these particles might be in the fine to ultrafine size and might therefore be 306 larger in number but lesser in weight as compared to those in the PM2.5 cut point collected from the 307 solid emission sources. 308 309

Elemental concentrations of emissions 310
The heavy metal concentrations in the emissions for the different burning simulations are 311 presented in Fig. 3 (accompanying values in Table S1). Presented in this table are the elements that 312 have acceptable recovery using the method reported by Rosales and Lamorena-Lim (2015). Of all 313 the elements detected, Pb appears to be in elevated concentrations, especially for MCC. It is not 314 certain what the Pb source is, although its presence in unequal amounts in MCC and CCP, both of 315 which used charcoal, may suggest that it came from the ingredients or the preparation of the chicken 316 used for simulations rather than the charcoal itself. P-values (compared against MCC mean of 15 317

A C C E P T E D M A N U S C R I P T
18 respectively, and thus Pb can be said to have come from the substrate cooked (i.e. chicken or its 319 seasonings). 320 Of the four fuel types, fuelwood (MFP) has the highest elemental emissions of Mn, Co, and 321 Ni. On the other hand, Cu and As are highest from the sawdust (KKP). Sr is highest in CCP and 322 Cd is almost the same for all four types, but slightly higher in CCP and MKP. For all samples, Ni 323 and Cu are seen in notable concentrations. Co is also observed in all samples except MCC. 324 These differences in concentrations may be explained by the differences in fuel sources. 325 The original environment of the different trees where the sawdust, charcoal, and fuelwood came 326 from might have been subjected to surroundings exposed to metal contamination. However, as 327 most of the fuels are from organic materials, heavy metals are not expected to be dominant in the 328 emissions. 329 330

Elemental Emission Factors 331
Tables 4 and 5 present elemental emission factors for MCC and four different fuel types. 332 Table 4 shows the emission factors in terms of time, while Table 5  The measured ion concentrations of all PM 2.5 samples collected from the five tested 351 materials are shown in Fig. 4 (accompanying values in Table S2). For MCC, CCP, KKP, and MFP, 352

A C C E P T E D M A N U S C R I P T
20 bound to organics in kerosene. Total ionic concentration for MFP was the lowest among all fuels, 355 while MKP and CCP show the highest total ion concentration. 356 Pearson correlation coefficients for the measured ions are shown in Table S6 (Westberg, 2003). In addition, the significant correlation between K + and 362 SO4 2for MCC is also attributable to biomass burning. As particles age, KCl particles are converted 363 to K2SO4 and KNO3 (Li et al., 2003). Considering that grilling lechon manok used charcoal, which 364 is prepared by pyrolysis, some KCl particles initially present may have aged and converted to 365 K2SO4 through time. However, since there is a lack of significant correlation between K + and SO4 2-366 for charcoal (CCP), this may also suggest that K2SO4 is produced during burning of the charcoal 367 itself. For MCC, burning was slightly longer (1 hour and 20 minutes) than for charcoal (1 hour). In 368 addition, the presence of the chicken may have contributed to the conversion process of the 369 potassium salts. 370 M A N U S C R I P T 21 Another notable correlation is that for Na + and NO3found uniquely in the MCC sample. 371 This correlation was not found in CCP, and thus is not a charcoal-related marker, but may be 372 characteristic of the chicken or its ingredients. 373 Significant correlations also exist between Na + and SO4 2-(0.96) and Na + and Cl -(0.70) in 374 MKP. This may be indicative of organic salts as possible kerosene additives. Similarly, a 375 correlation was found for Na + and SO4 2-(0.75) and Na + and NH4 + (0.77) in KKP, for which a little 376 bit of kerosene was used to start the burning process. One possible source of these ions are antistatic 377 additives, such as salts of organic acids and quaternary ammonium salts, that are used to increase 378 the electrical conductivity of hydrocarbons such as gasoline and kerosene (Wauquier 1995). 379 For MFP, various ion pairs (other than K + and Cl -) such as Mg 2+ and SO4 2-(0.71), Ca 2+ and 380 Cl -(0.89), Ca 2+ and SO4 2-(0.67), and Ca 2+ and NO3were found. In addition to the fuel type itself, 381 these ion pairs could be attributed to the chicken itself or the ingredients used to flavor the chicken. 382 383

Water-soluble organic carbon (WSOC) 384
Aside from the ionic composition, organic carbon is one of the major components of water-385 soluble components of atmospheric aerosols (Ram and Sarin, 2010; Sannigrahi et al., 2006). Figure  386

A C C E P T E D M A N U S C R I P T
22 hydrocarbons, it should be noted that these are long-chain (aliphatic) and cyclic hydrocarbons (The 389 Editors of Encyclopaedia Britannica, 2016) and an incomplete combustion may result to water-390 insoluble organic compounds. 391 It can be noted that for biomass, the order of WSOC follows that of the particulate emission 392 factors (i.e. MFP < CCP < KKP). MFP is also consistently low in water-soluble components (Figs. 393 4 and 5). In addition, while charcoal, sawdust, and fuelwood are all derived from biomass, it is 394 interesting to note that the WSOC range of these fuels span two orders of magnitude. This range, 395 as well as the order, may be attributed to differences not only in chemical makeup but also from 396 the physical form (size)-sawdust was very fine, while charcoal and fuelwood were both bigger 397 pieces and thus had less surface area, making its composition less accessible. Burning very fine 398 sawdust, which had the greatest surface area, led to more efficient burning and possibly more 399 complete combustion compared to the other two biomass fuels. The more complete the combustion, 400 the higher the composition of oxygenated C functional groups (e.g. COOH, aldehydes, ketones, 401 etc.) likely to exist (Sannigrahi, 2006). However, information on the speciation of the organic 402 portion is sparse and can only be inferred indirectly from WSOC analyses. 403

A C C E P T E D M A N U S C R I P T
23 Biomass burning emissions may be detected in atmospheric PM using levoglucosan and 406 other related monosaccharide anhydrides (mannosan and galactosan) derived from the breakdown 407 (dehydration) of cellulose, a major component of wood (Simoneit et al., 1999). As such, these 408 sugars are used as molecular markers for combustion of vegetation (Kuo et al., 2008). In addition, 409 these sugars, when correlated to K + , can be used as tracers for biomass burning (Jung et al., 2014). 410 The levels of monosugars detected in each fuel type is shown in Fig. 6 and Table S3 2008). Since charcoal has already undergone pyrolysis during its preparation, it is expected that 420 H/C ratio would be lower (i.e. most of it has been converted to black carbon or soot). As such, 421 sugars were found in low levels in CCP. However, MCC, which also used charcoal, is shown to 422 have about an order of magnitude total sugars compared to CCP. While these sugars may not come

A C C E P T E D M A N U S C R I P T
24 from the charcoal itself, they may have originated from the cooked substrate (i.e. chicken and spices 424 stuffed inside such as lemongrass) instead. 425 To further explore the relationship between each sugar as well as with other ions, a 426 correlation analysis was run (Tables S5 and S6). No significant correlations were observed for CCP. 427 As for MKP, which is not biomass, levoglucosan and galactosan were not detected in any of the 428 samples, and mannosan was detected in only three samples. Levoglucosan is highly correlated with 429 mannosan, with Pearson coefficients ranging from 0.96-0.99 for MCC, KKP, and MFP. 430 Levoglucosan is also highly correlated with galactosan for the fresher wood samples (sawdust and 431 fuelwood), but correlation is lower between the two for MCC. Similarly, galactosan and mannosan 432 are highly correlated for KKP and MFP but has a lower correlation coefficient for MCC. 433 Nevertheless, these values confirm that the emission of these three sugars are related to burning 434 vegetation. However, mannosan and galactosan were not found in samples of charcoal tested. 435 Moreover, for the CCP PM samples where the sugars were found, correlation was not statistically 436 significant. Since charcoal is burned wood, cellulose have already been broken down prior to use 437 in cooking. However, for the MCC samples, the chicken was stuffed with lemongrass. The burning 438 of the lemongrass leaves produced fresh cellulose breakdown products. Of all the samples, sawdust 439 emitted the highest concentrations of all sugars. Since it was introduced as fine wood shavings,

25
The potassium cation (K + ) has been previously reported in literature as a conventional 442 biomass burning tracer (Mochida et al., 2010) and a major constituent of biomass ash (Schmidl et 443 al. 2008). It has also been found to be moderately correlated with levoglucosan, mannosan, and 444 galactosan (Jung et al., 2014) in biomass. However, it may not exclusively come from biomass-445 burning-for example, sea-salt is another source of K + (Mkoma et al. 2013). Thus, its ratio with 446 any of the anhydrosugars (levoglucosan, mannosan, or galactosan) that are known to be biomass-447 burning tracers themselves, e.g. Levo/K + , is an effective tool as a biomass burning tracer and can 448 be used with confidence (Jung et al., 2014). (ca. 14-15) are usually observed for hardwood, while low ratios (ca. 3.6-3.9) are usually observed 453 for softwood (Schmidl et al. 2008). Table 6 shows the average ratios obtained. Levo/Man ratios 454 show a very small standard deviation for KKP, which means that levoglucosan and mannosan are 455 emitted in proportional amounts during combustion and may thus be used as a tracer for red lauan 456 sawdust/wood shavings. As for chicken and fuelwood burning, the Levo/Man ratio also presents a 457 potential marker. However, the value obtained for the chicken grilling may be misleading because 458 the value obtained in this study considers the presence of the lemongrass stuffing. Moreover, since

A C C E P T E D M A N U S C R I P T
26 there are many types of fuelwood that may be used for household cooking, having a range of 460 Levo/Man ratios that is inclusive of more wood types will be beneficial as chemical fingerprint of 461 biomass burning. On the other hand, a K + -sugar correlation was not existent in any of the samples 462 in this study. It might be that the relationship is not linear, due to differences in the time it requires 463 for K + and the sugars to settle on the PM emitted. The lack of correlation may also be due to the 464 specificity of the abundance of K + from certain biomass types. Moreover, Levo/K + ratios show 465 large standard deviations for the ratios calculated herein; thus, this ratio may not be consistent 466 throughout the burning of the fuel and may not be the best choice as a marker for the types of fuels 467 well as the ratio of levoglucosan to mannosan, as chemical fingerprints for biomass burning in local 496 urban areas, including indoor PM samples. Specific ion pairs also provided more understanding

A C C E P T E D M A N U S C R I P T
28 into possible additives in processed fuel, as well as ion pairs unique to the cooked substrate. WSOC 498 analysis provided insight on the correlation of biomass size to the transfer of water-soluble organic 499 combustion products to the particle phase, which paves a pathway to introduce various pollutants 500 in the human respiratory system. Charcoal-related emission factors were found to be lower than 501 literature values, which may result to underestimations of emissions when the emission factors are 502 obtained from these references. It was also found that processed fuels (charcoal and kerosene) vary 503 in their sugar content as opposed to "fresher" (unprocessed) fuels (fuelwood and kusot) as a result 504 of the industrial process that fuels undergo which leads to a lower H:C ratio. It is also noted that 505 sugars are also dependent on the substrate cooked. Therefore, this must be considered when 506 attributing compositions to fuel sources. Recommendations for future studies include (1) more test 507 materials applicable to the local or regional setting, (2) speciation of the water-soluble organic 508 carbon in the particle phase, in order to better characterize and estimate the gas-particle partitioning 509 and particle growth, and (3) estimation of dose for different members of the population.

A C C E P T E D M A N U S C R I P T
37

A C C E P T E D M A N U S C R I P T
41