Monitoring and modeling of household air quality related to use of different Cookfuels in Paraguay

Abstract In Paraguay, 49% of the population depends on biomass (wood and charcoal) for cooking. Residential biomass burning is a major source of fine particulate matter (PM 2.5) and carbon monoxide (CO) in and around the household environment. In July 2016, cross‐sectional household air pollution sampling was conducted in 80 households in rural Paraguay. Time‐integrated samples (24 hours) of PM 2.5 and continuous CO concentrations were measured in kitchens that used wood, charcoal, liquefied petroleum gas (LPG), or electricity to cook. Qualitative and quantitative household‐level variables were captured using questionnaires. The average PM 2.5 concentration (μg/m3) was higher in kitchens that burned wood (741.7 ± 546.4) and charcoal (107.0 ± 68.6) than in kitchens where LPG (52.3 ± 18.9) or electricity (52.0 ± 14.8) was used. Likewise, the average CO concentration (ppm) was higher in kitchens that used wood (19.4 ± 12.6) and charcoal (7.6 ± 6.5) than in those that used LPG (0.5 ± 0.6) or electricity (0.4 ± 0.6). Multivariable linear regression was conducted to generate predictive models for indoor PM 2.5 and CO concentrations (predicted R 2 = 0.837 and 0.822, respectively). This study provides baseline indoor air quality data for Paraguay and presents a multivariate statistical approach that could be used in future research and intervention programs.


| 253
TAGLE ET AL. exposure can lead to harmful neurological effects. 7 The health issues of prolonged exposure to PM 2.5 from cooking with biomass have been associated with a higher risk of suffering pneumonia in children and cardiovascular and pulmonary diseases in adults. 8,9 Biomass-burning cookstoves have the potential to produce indoor air pollution when used in poorly ventilated household environments. As an environmental factor, household air pollution has been associated with an increased risk of premature death. 10 Due to the threat it poses to public health, the World Health Organization (WHO) has established guidelines on indoor air quality exclusively related to household fuel combustion. 11 These recommend that indoor PM 2.5 concentration should not exceed 10 μg/m 3 as an annual average, while the daily CO average should be below the threshold of 7 mg/m 3 , approximately 5.7 parts per million (ppm).
Despite WHO suggestions, healthy indoor air levels may be difficult to achieve in countries where biomass is in high demand for household energy needs, as in present-day Paraguay. In Latin America, Paraguay has one of the highest percentages of population dependent on biomass as the main fuel used for cooking (49%), after Haiti (91%), Guatemala (57%), Nicaragua (54%), and Honduras (51%). 2 In addition, excessive consumption of biomass for energy production has helped sustain progressive deforestation in the country. 1,12 The present study characterizes and models indoor air pollution related to biomass burning and low-emission cookstoves in response to the growing need to know the national state of indoor air quality, especially in rural areas where wood and charcoal are used by the majority of households. This research was conducted in collaboration with major stakeholders involved with environmental health in Paraguay: the Pan American Health Organization (PAHO) and the Dirección General de Salud Ambiental (DIGESA). The measurements and analyses presented in this article provide a foundation for establishing a baseline that could be used in future studies, as well as in potential cookstove intervention projects.

| Study site
The study was conducted in July 2016 (winter) in two low-income

Practical Implications
• Household air pollution associated with cooking fuels has been well documented in various developing countries but not in Paraguay.
• The paper reports the first indoor air quality monitoring campaign conducted in the country.
• These data could be used to model indoor air quality in similar settings and to develop national policies aiming to reduce exposure to household air pollution.

| Household selection
In June 2016, survey data about fuels used for cooking, heating, and lighting were collected in 238 rural households at JAS and LIM. The survey was designed and administered by PAHO based on WHO's World Health Survey. A database was created without personal identifiers but including household information, such as location and the type of fuel used for cooking (wood, charcoal, LPG, and electricity). Households were stratified according to the type of fuel used for cooking, and, in each subset, households were randomly selected to be visited each day for conducting measurements. The field team introduced the study and its measurements to the head of the family, who was invited to participate. Formal recruitment occurred after informed consent was obtained from the head of the family.
Households with pregnant women or smokers were excluded.

| Indoor air monitoring
Household air pollution was monitored in the cooking area for 24 hours. A sample deployment is shown in Figure 2. Sampling was performed on weekdays, starting one morning (8-9 AM) and ending the subsequent morning. PM 2.5 and CO monitors were colocated approximately 1.5 m away from the cookstove and at adult breathing height (1.6 m above the floor). PM 2.5 was collected on a pre-weighed PTFE (polytetrafluoroethylene) filters (Pall Corporation, NY, USA) for posterior gravimetric analysis (37 mm, 2.0 μm pore size). Filters were placed inside a threepiece cassette (23370-U, Sigma-Aldrich) and backed with a drain disk (36 mm, Whatman ® ). Cassettes were coupled to a PM 2.5 cyclone sampler (Triplex SCC1.062, Mesa Labs, USA) and connected to an air pump (AirChek XR5000, SKC, USA). The initial flowrate was adjusted to 1.5 L/min. At the beginning and end of the sampling event, the flowrate was measured using a digital flowmeter (Challenger CH100 flowmeter, Mesa Labs). To estimate the volume of air sampled (m 3 ), the average of the pre-and post-sampling airflow rates (m 3 /min) was multiplied by the total sampling minutes displayed on the pump screen.
For gravimetric analysis, filters were weighed before and after

| Outdoor air monitoring
In order to determine the PM 2.5 concentration outdoors, a central location in each village was selected for installing a fixed monitoring station. The equipment was placed on the roof of households that only used electricity for cooking, approximately 2.5 m above the ground and away from direct emissions of any kind. Time-integrated (24 hours) PM 2.5 samples were collected on 37-mm PTFE filters using a two-stage impactor (4 L/min) described elsewhere. 13 Sampled filters F I G U R E 2 Household air pollution monitors colocated in the kitchen area. CO monitor (1), triplex cyclone for PM 2.5 collection (2), and air temperature sensor (3) were subjected to the gravimetric analysis described in Section 2.3 and to X-ray fluorescence (XRF) spectrometry to quantify concentrations of elements ranging in atomic number from 11 (Na) to 82 (Pb).
The XRF spectrometry was performed with the Epsilon 5 spectrome- To determine the daily profile, the one-minute PM 2.5 concentrations were averaged over one hour. Meteorological parameters, such as hourly wind direction, wind speed, precipitation, and temperature, were obtained from the Agricultural Science Department of the National University of Asunción. The faculty operates a meteorological station located 13 and 18 km from JAS and LIM, respectively (25°20′0.04″S, 57°31′0.02″W).

| Predictor variables
During the monitoring campaign, potential predictors of PM 2.5 and CO concentrations were recorded as either categorical or continuous variables. A structured questionnaire was applied at both the beginning and end of the monitoring session to capture several variables at the kitchen level, as shown in Table 1. Variables included the rural community; main fuel used for cooking; the construction materials of the roof, floor, and walls; kitchen structure; occurrence of sweeping, heating, and smoking; as well as burning of incense, mosquito repellent (indoors), and garbage (outdoors). Communitylevel statistics are presented in Table S1.
The parameters recorded as continuous variables are shown in Table 2. Cookstove usage was monitored for 24 hours using a temperature sensor (iButton DS-1922T, Maxim Integrated, CA, USA) as a stove use monitor (SUM). 14,15 These sensors were attached with tape to the base of cooking appliances and recorded temperature every 1 minute (T stove ). Temperature inside the kitchen (T air ) was recorded using a HOBO datalogger (Onset Inc, USA) colocated with the air samplers. The total time of cookstove usage was the sum of minutes in which the T stove was at least 10°C above T air . For measuring the kitchen room volume and distance between the air samplers and the cookstove, a laser length meter was used (GLM 40, Bosch).

| Data analysis
Descriptive statistics are presented as means, standard deviations (SD), and confidence intervals (CI) of the mean. To determine groups that were significantly different from each other, an analysis of variance (ANOVA) followed by Tukey's multiple comparison test was performed (significant at P-value < 0.05).
Predictive models were created from the observed indoor concentrations and potential explanatory variables shown in Tables 1   and 2. As a preliminary step, the normality of the distribution of The predicted R 2 in the final model was estimated using the "olsrr" package. The PCA for outdoor samples was performed with the "factoextra" and "corrplot" packages.

| Cookstoves and kitchen features
The monitored households used one of four cooking methods: three-stone open fires for burning wood, metal braziers for burning charcoal, regular LPG cookstoves, or electric hot-plate cookers (shown in Figure S1). None of the households used more than one type of cooking method during the measurement period.
Households had one of two kitchen configurations: enclosed (a kitchen inside or next to the household with a roof and four walls) and semi-enclosed (a cooking room with a roof and three walls).
Semi-enclosed kitchens were not found in households that cooked using LPG or electricity.

| Indoor PM 2.5 and CO concentrations
The 24-hour average indoor PM 2.5 and CO concentrations observed in different kitchen and fuel settings are summarized in Table 3. The highest average (±SD) PM 2.5 concentrations were observed in woodburning kitchens, specifically in the enclosed type (851 ± 656 μg/ m 3 ). Those kitchens using the same fuel but with a semi-enclosed structure had a lower average PM 2.5 concentration (681 ± 95 μg/m 3 ).
CO concentrations were similar between kitchen configurations.
In total, the enclosed and semi-enclosed charcoal-burning kitchens had average PM 2.5 concentrations of 107 ± 69 μg/m 3 and CO concentrations of 7.6 ± 6.5 ppm. Both pollutants were observed at higher concentrations in enclosed structures.
The kitchens using LPG and electricity had the lowest average concentrations of both PM 2.5 and CO. Despite this, the average PM 2.5 concentrations in these kitchens were 52 ± 17 μg/m 3 , higher than the values expected for an emission-free environment.

| Outdoor PM 2.5 and meteorology
The 24-hour average outdoor PM 2.5 concentration was 27.5 μg/ The major element found in the outdoor PM 2.5 was potassium (K), a chemical tracer associated with biomass and agricultural burning. 16 The mass of all the elements reported by the XRF spectrometry (Table S3) (Table S4).

| Indoor PM 2.5 predictive model
Through the statistical procedure described in Section 2.6, a five-     Table 5 and Figure 6, respectively. The burning of wood and charcoal, indoor PM 2.5 concentration, and cookstove usage were variables significantly associated with higher CO concentrations. Similar to the predictive model for PM 2.5 , outdoor garbage burning was the only external variable significantly associated with increased CO concentration in the kitchen (P-value = 0.006). In a bivariate model ( Figure S2), both pollutants also presented a significant association, although the correlation was lower (R 2 = 0.63).
The cross-validation resulted in a relatively high adjusted R 2 (0.762 ± 0.037) and a low RMSE (0.365 ± 0.033). Even though the selected model fits well, the intercept had a significant association, which indicates the existence of other unincorporated variables that may also contribute to the variation in the CO indoor concentration.

| D ISCUSS I ON
To the best of our knowledge, indoor and outdoor air quality data have not been previously reported in the scientific literature for Paraguay. Based on the results of this study, the 24-hour average PM 2.5 concentrations in both indoor and outdoor environments exceeded the guidelines established by the WHO (35 μg/m 3 , Interim-Target 1). We suggest that outdoor PM 2.5 concentrations in the rural communities could be strongly influenced by biomass burning for cooking and waste burning.  26 and Guatemala (6.8 h/d). 27 As shown in Table 6, a similar magnitude of PM 2.5 and CO concentrations has also been reported in rural households in Nepal, 28,29 Pakistan, 30 and China. 31 As observed in Paraguay, concentrations of PM 2.5 higher than the values expected were recorded in households  from households using solid fuels (biomass and coal) have been indicated as the main factor responsible for increasing PM 2.5 inside homes using clean fuels. 32,33 In our study, the large contribution of K found in ambient PM 2.5 suggests that outdoor air quality was considerably impacted by biomass burning. F I G U R E 6 Goodness of fit of the predictive model for LnCO. The blue line is model fit; the red line is a 1:1 line concentrations of both pollutants were observed in kitchens that used LPG or electricity; however, these kitchens had higher-thanexpected PM 2.5 concentrations; this could be associated with external sources, such as burning of biomass and garbage in community spaces. Two regression models were developed to estimate indoor PM 2.5 and CO concentrations, which have a predictive power of over 85%. Both models can be considered when designing national cookstove intervention projects, as well as in cost-benefit analysis.

ACK N OWLED G EM ENTS
We gratefully acknowledge the families who kindly agreed to participate in this study. Also, we express our thanks to the local support given by DIGESA, especially to the Chief Director Luis Leguizamón and the Head of Air Quality Laura Flores. We also thank field assistants Alice Bergottini and Claudia Acosta for their significant con-