Familial Differences in Personal PM_2.5 Exposure Within a Rural African Community Explained with Spatiotemporal Exposure Apportionment

Abstract

Exposure to fine particulate matter (PM_2.5) from solid fuel combustion is a major determinant of global morbidity and mortality. However, variations in exposure remain uncertain across many high-risk populations. This work describes personal PM_2.5 exposures among household members (adult men, adult women, and children) in rural sub-Saharan Africa, where biomass fuel is the primary household energy source. We assessed personal PM_2.5 exposures using wearable monitors that combined real-time sensing, time-integrated (gravimetric filter) sampling, and continuous location-activity tracking over 48-hr periods. A total of 1,280 samples were collected from 579 Rwandan homes over a 15-month period comprising: 304 men (aged 23–84 yr), 495 women (aged 20–84 yr), and 481 children (aged 8–17 yr). Linear mixed models, controlling for household, suggested that children were exposed to 14% (CI: 6, 22%) more PM_2.5 than their mothers and 100% (CI: 85, 117%) more than their fathers. Spatiotemporal analyses, aggregated into various microenvironments (e.g., home, school, transit, agricultural fieldwork), reveal that children bore a disproportionate exposure burden from in-home cooking activities compared with their parents. Results from this work indicate that interventions for household energy systems, in conjunction with familial lifestyle and behavior modifications, are necessary to reduce personal PM_2.5 exposures in rural Rwanda, especially among children.

Graphical Abstract (TOC Art)

Created by the Authours.

Introduction

Exposure to PM_2.5 (particulate matter with an aerodynamic diameter less than 2.5 μm) was responsible for an estimated 7.8 million premature deaths and 231 million disability-adjusted life years (DALYs) lost globally in 2021.^1,2 Health outcomes attributed to PM_2.5 exposure include incidence and exacerbation of chronic obstructive pulmonary disease, asthma, ischemic heart disease, acute lower respiratory tract infections, stroke, and lung cancer, among others.^3–7 The PM_2.5 disease burden, however, remains uncertain for many demographic groups and across various geographic regions due to a lack of collected personal PM_2.5 exposure data. For example, a recent review of 140 personal exposure studies that met established quality-control criteria (having at least 10 unique participants, a minimum of 20 hours of continuous monitoring, and the use of the same type of monitoring equipment throughout the duration of the study) found that none of these studies compared personal PM_2.5 exposure among men, women, and children in either low-income or low-middle-income countries (LICs and LMICs).⁸ To date, most personal PM_2.5 exposure studies in sub-Saharan Africa have focused on women and/or young children (0 to 5 years), as these groups are assumed most likely to have the closest proximity to cooking emissions.^9–14 Exposure data are lacking for men and children (5 to 17 years) in these settings; these data are necessary to define exposure burden more accurately as a function of population demographics.

Personal exposure (i.e., data collected from within one’s breathing zone) is considered the “gold standard” for estimating individual risks because it accounts for variations in exposure that occur as a function of time, location, activity, and behavior. Personal exposure data are especially important for risk assessment differentiation between various demographic groups within the same geographical region.^8,15,16 When personal exposure data are lacking, mortality and morbidity burden estimates rely on a combination of satellite imagery, chemical transport modeling, regional source-receptor estimates, and fixed-site outdoor concentration measurements (which can be several km away from many at risk populations).^8,15,17–21 These methods are commonly used to estimate exposure-response relationships for ambient PM_2.5, but they are less reliable at quantifying indoor air pollution exposure, which is characterized by more spatial and temporal variation.^9,22,23 Understanding individual PM_2.5 exposure patterns in sub-Saharan Africa is crucial since this region accounts for an estimated 31% of global deaths related to household air pollution, despite having only 15% of the global population in 2021.^1,24 Unfortunately, personal PM_2.5 exposure studies are often limited due to a lack of wearable, robust, accurate, and readily deployable (with standardized quality assurance measures) PM_2.5 monitoring technology.¹⁶ However, recent improvements in personal PM_2.5 monitoring technology have increased our ability to track exposures in space and time with less disruption to individual’s daily routines.^9,22,25–32

The primary objectives of this work were to quantify the PM_2.5 exposures and demographic variations among adults (both men and women) and adolescent/pre-adolescent children (aged between 8 and 17 at the time of enrollment, hereafter referred to as “child” or “children”) in a rural sub-Saharan African community that relied on solid biomass fuel as their sole form of household cooking energy. We combined space-and-time-resolved personal exposure data, a machine-learning cluster analysis, and geocoded land-use feature data to apportion personal PM_2.5 exposures into unique time-activity patterns. These patterns helped reveal source-receptor relationships for PM_2.5 exposure that explain demographic variations in daily exposures among men, women, and children in this rural setting.

Methods

Study Setting and Participant Eligibility

The SHEAR project (Sustainable Household Energy Adoption in Rwanda) is a 3-year randomized control trial (2022–2025; US ID: NCT05668624) located in the communities of Karambi and Isangano within the Eastern Province of Rwanda (see Figure 1 for a map of the study region). The villages were chosen because they will not be receiving distributed electricity during the study period (there is a plan to connect them to the national grid eventually) and solid fuels for cooking are used nearly universally amongst the households in the study area. These villages typify rural sub-Saharan African subsistence farming communities with limited resources and virtually no access to modern forms of household energy (e.g. grid electricity or piped natural gas). Data presented here were collected during the SHEAR baseline period, prior to the intervention that introduced liquefied petroleum gas for cooking and solar power for lighting and other basic needs. A total of 630 households were initially enrolled in SHEAR with the following inclusion criteria: (1) located in the study area of Isangano and Karambi; (2) had an adult male and female, or a single adult, head of household over 18 years, and at least one child 8 to 17 years of age; (3) non-smoking; (4) primarily used an open fire or wood/charcoal stove; (5) did not do any commercial cooking; (6) were not currently pregnant (women); and (7) did not plan to move within the next 3 years. Participants provided informed consent approved by the Rwanda National Ethics Committee (# 00001497) and by the Institutional Review Board at Colorado State University (IRB # 3946). Additional details on the SHEAR study design, enrollment criteria, household characteristics, and participant surveys are provided in the online supplement.

Figure 1.

Map of the Study region. Green dots represent the locations of SHEAR study households (n = 579), blue dots depict the AMOD deployment locations (ambient air monitoring sensors n = 4), the yellow polygons are various fields participants conduct agricultural work in, and the red polygons show the school boundaries. The inset map of Africa in the top left corner of the figure highlights Rwanda, the country where this study is located and the blue dot within this inset map shows the exact location of the SHEAR study. The bottom left insert image depicts a typical roadside in one of the study villages.

Personal PM_2.5 Exposure

Personal PM_2.5 exposure sampling consisted of wearing a UPAS V2 Plus (Ultrasonic Personal Aerosol Sampler; Access Sensor Technologies, Fort Collins CO) in the breathing zone for a continuous 48-hour period, during which participants went about their daily routines.²⁵ The UPAS is a wearable air quality monitor that uses a quiet piezoelectric pump to draw air at 1 L∙min⁻¹ through a 2.5 μm size-selective cyclone inlet and onto a 37 mm PTFE sampling filter (PT37 MTL LLC, Minneapolis, MN)³³. A separate stream pulls air into a light-scattering sensor (SPS30, Sensirion AG Stäfa, Switzerland) at the front of the unit to estimate PM_2.5 concentration at 30-second intervals.³⁴ The UPAS V2 Plus also records GPS, accelerometry, air temperature, relative humidity, and atmospheric pressure data at the same 30-second intervals. Air sampling filters were pre- and post-weighed in triplicate (XS3DU, Mettler-Toledo, Columbus, OH, USA) inside a climate controlled environment via an automated robotic process at Colorado State University and shipped to and from the field via air freight every other month.³⁵ Filters were stored in an on-site freezer (−20 °C) after sample collection between the bi-monthly air freight shipments. Post-sampling filters were stored in a −80 °C freezer once received by CSU before equilibration in the climate-controlled weighing chamber prior to gravimetric analysis. Filter field blanks were collected at the beginning of each sampling day and used to account for potential measurement artifacts encountered during shipping and handling.

Participants were asked to wear the UPAS on their chest during waking hours, via a custom harness (Figure S2) produced by a local tailor. To obtain a continuous 48-hour sample, a small battery-powered charging station was provided to each participant to recharge their UPAS overnight (Figure S3). Participants were instructed to keep the UPAS and charging station as close to their breathing zone as feasible while they slept (see online supplement for additional details). Participants could also remove the UPAS and keep it nearby when bathing or using the restroom. Device wearing compliance was assessed by UPAS accelerometry data, evaluated between the hours of 6:00 and 21:00 (see the UPAS Wearing Compliance section in the SI for more details). Data presented here are from the SHEAR baseline survey of study participants; each sample represents a single 48-hr measurement (n=1,280 samples total with no repeated measures).

Monitor deployment, collection, in-field filter handling, and participant interactions were conducted by trained local staff. Integrating staff from the community improved the project’s rapport among participants and helped establish trust within the larger community, as most participants only spoke Kinyarwanda.

Microenvironmental Exposure Apportionment

Each 30-second PM_2.5 observation was apportioned into one of 8 distinct microenvironments based upon its GPS coordinates: home, near home, agricultural field, school, in transit (within village), in transit (outside of village), other, and no-GPS. We utilized a machine-learning clustering algorithm (dbscan from the fpc package using R version 4.2.2)³⁶, in conjunction with a matrix of geocoded boundaries, to apportion each 30-second PM_2.5 observation to a unique microenvironment. Unreliable GPS data were culled using a cleaning algorithm, which set the latitude and longitude values for erroneous observations to “NA”. All GPS observations that were culled or otherwise missing were assigned to the “no GPS” microenvironment. See the online supplement for more details about the GPS cleaning steps, the machine learning clustering algorithm, and the complete microenvironmental assignment process.

Ambient PM_2.5 Monitoring

Ambient PM_2.5 levels were routinely monitored using 4 Aerosol Mass and Optical Depth monitors (AMODs, developed at Colorado State University) set upon 4m (13.1 ft) poles at two central and two edge locations in Karambi, shown as the blue dots in Figure 1.^37–40 These ambient air quality monitors functioned similarly to the UPAS personal monitors and were corrected using co-located gravimetric sampling. Additional details on the ambient air quality monitors, their filter sampling schedule, the rationale behind their monitoring height, and correction ratio calculations can be found in the Ambient Air Quality Monitoring section of the online supplement.

Data Management and Statistical Analyses

Gravimetric filter data were used to correct the 30-second PM_2.5 concentration data on a per-sample basis using methods outlined in Tryner et al. 2020 (see the Gravimetric Correction for Realtime PM_2.5 Estimates section in online supplement for details).⁴¹ This correction was necessary because low-cost light scattering PM_2.5 sensors are sensitive to changes in aerosol size, refractive index, and ambient relative humidity – all of which can vary during sampling.⁴² Personal PM_2.5 samples were included in analysis if they had at least 36 hours of recorded run-time (a full analysis of how we calculated this minimum runtime using a bootstrapping analysis can be found in the SI document), a gravimetric correction ratio within 2 geometric standard deviations of the geometric mean of all the sample correction ratios, and if their recorded accelerometry data exceeded movement thresholds during waking hours; see the Valid Personal PM_2.5 Exposure Sample Criteria section in the SI documentation for more details. Aggregate exposure measures calculated for each individual were 48-hr time-integrated concentrations. Diurnal trends were assessed with hourly averaged PM_2.5 concentrations. Microenvironmental apportionment analyses utilized 30-second PM_2.5 concentrations. All the PM_2.5 concentrations used in our analyses were corrected to the PM_2.5 filter sample loading, collected concurrently.

Linear mixed models were developed to examine the statistical significance, at a type 1 error rate of 5%, for the differences in personal exposures among men, women, and children while accounting for household using random intercepts. PM_2.5 exposure concentrations were natural log-transformed to meet the model assumptions (reducing right-skewedness and heteroscedasticity), taking the form of:

$$Y_{ij}={\beta }_0+{\mathit{\beta }}^T{\mathit{X}}_{\mathit{i}\mathit{j}}+{\alpha }_i+{ϵ}_{ij}$$

Where $Y_{ij}$ is the log-transformed average PM_2.5 exposure concentration from the ith household and the jth participant, ${\mathit{X}}_{\mathit{i}\mathit{j}}$ is the set of fixed predictor variables, $\mathit{\beta }$ is the vector of coefficients for the predictor variables, ${\alpha }_i$ is a random intercept term for household $i$ and ${ϵ}_{ij}$ represents the residual model error (assumed to be normally distributed with mean zero and constant variance). See Tables S3–7 and S9–10 in the SI for details on model fixed predictor variables (e.g. demographics and time partitioning) and results.

Results and Discussion

Our analysis included a total of 1,280 valid samples, 304 from men, 495 from women, and 481 from children. Of the initial 1,479 personal exposure samples collected, 139 (9%) were removed due to inadequate runtime, 46 (3%) were removed for poor wearing compliance, and 53 (3.5%) were flagged for having filter-to-sensor PM_2.5 correction ratios beyond 2 geometric standard deviations from the geometric sample mean (see the Valid Personal PM_2.5 Exposure Sample Criteria section, Table S2, in the SI for more details). Men, women, and children had similar percentages of valid samples, 80.9, 79.50, and 75.3%, respectively. Overall, children had the highest 48-hour time-averaged PM_2.5 exposures with a median concentration of 193 μg∙m⁻³. Women had the second highest exposures with a median concentration of 174 μg∙m⁻³, and men had the lowest exposures with a median of 94 μg∙m⁻³ (Figure 2, Table 1). These exposure levels greatly exceeded the WHO annual average interim 1 target (IT-1) of 35 μg∙m⁻³.⁴³ This guideline was chosen as an optimistic milestone for this study population since the median daily average ambient PM_2.5 concentration observed during this baseline period was 23.0 μg∙m⁻³. The mean ambient PM_2.5 concentration for these study villages was 27.0 μg∙m⁻³, which falls below the WHO IT-1 target of 35 μg∙m⁻³; however personal exposures greatly exceeded that target (Figure 2, Table 1). This finding demonstrates that even regions with low ambient PM_2.5 levels can still have population exposures far exceeding health-based guidelines. A Wilcoxon signed rank test suggested the difference between children’s and women’s exposures within the household was distinct from 0 μg∙m⁻³ (CI: −46.6, −19.0 μg∙m⁻³). The same test suggested children’s exposures were even more distinct from the men’s within their household (CI: −103.8, −71.5 μg∙m⁻³). Descriptive statistics for the 48-hour personal PM_2.5 exposure data are provided in Table 1.

Figure 2.

Boxplots represent the average (~48-hr for personal samples, and 24-hr for ambient) PM_2.5 concentration distribution among ambient air, and personal exposure samples of men, women, and children. The Y-axis is logscale. The ambient box depicts the distribution of all the daily average PM_2.5 concentrations recorded across all sampling locations within the same 15-month window when personal exposure sampling occurred. Center lines delineate the median; boxes represent the interquartile range, and whiskers represent the extent of observations within +/− 1.5·IQR. The red line is the WHO interim 1 target of 35 of μg∙m⁻³. The black diamonds represent the means.

Table 1.

Descriptive statistics for 48-hour average PM_2.5 concentrations (μg∙m⁻³) as a function of familial demographics. IQR = Inter-quartile range.

		48-hour Sample Average PM2.5 Concentration (μg·m−3)
	Sample Size (n)	Mean	Standard Deviation	Geometric Mean	Geometric SDa	Median	Min	Max	IQR
Children	481	248	177	199	1.96	193	31	1,235	214
Women	495	227	233	174	2.01	174	28	3,551	173
Men	304	127	101	98	2.06	94	16	739	104
All	1,280	212	194	160	2.11	158	16	3,551	170

^(a)Geometric standard deviation is dimensionless.

Linear mixed modeling (n = 1280 participants in 579 unique households) accounting for household as a random effect suggests children were exposed to 14% (CI: 6, 22%) higher PM_2.5 concentrations than women in their household and 100% (CI: 85, 117%) higher concentrations than the men (See Table S3 for details). All participant exposures were higher than the median ambient air concentration, 23.0 μg∙m⁻³, indicating personal exposures were dominated by proximity to localized PM sources. These demographic exposure differences, along with ambient PM_2.5 concentrations, are illustrated in Figure 2.

The finding that children had significantly higher PM_2.5 exposure levels than their mothers was unexpected, given prior reports of household air pollution exposures in Africa.^8,10,12 Children are particularly vulnerable to the effects of air pollution due to their developing respiratory and immune systems; yet, their primary guardians (often assumed to be their mother) are frequently used as a proxy for their exposure due to assumed similarities in daily activity patterns.^44,45 Based on participant survey responses, women did the majority of the household cooking tasks; 65% of women report being the “Primary Cook” and 39% report being “Both the primary cook and head of my household”. Cooking was the only major source of PM_2.5 within the community (vehicular traffic and trash burning were infrequent), so it was notable that children had the highest levels of PM_2.5 exposure among the demographics. A prior study of personal PM_2.5 exposures in rural Africa reported that Ugandan women had 51% and 574% (60 and 151 μg∙m⁻³) higher 24-hour average exposures than their female and male children respectively (aged 6 – 17 years old); Ethiopian women had 53% and 578% (71 and 175 μg∙m⁻³) higher average exposures than female and male children respectively.¹⁰

The children’s PM_2.5 exposures reported here were also higher than children from LMIC and LICs as reviewed by Lim et al. (2022). Children from our study had a median PM_2.5 exposure of 193 μg∙m⁻³ compared to 65 μg∙m⁻³ reported for the groups of children from other LMICs and LICs (n = 15 studies).⁸ Further, Lim et al. reported that all of the adult groups (n=26) from LMICs and LICs had median PM_2.5 exposures of 79.5 μg∙m^−3. These adults groups were not linked to the children groups (different samples from different studies), but across all the studies reviewed the children had lower median PM_2.5 exposures.⁸ Our results have implications for mortality and morbidity estimates from household air pollution for low- and middle-income countries around the world.^1,7,46–49 The elevated exposure ratio experienced by children relative to women contradicts common assumptions made by the global burden of disease study and highlights the need for additional research examining demographic variation in personal PM_2.5 exposure.⁴⁷

For our study, personal PM_2.5 exposures between male and female children were not significantly different. Linear mixed modeling suggested that boys have only 3% higher PM_2.5 exposures (CI: −10 – 16%) than girls (see table S4 for details) and that sex was only a significant predictor of exposure differences for adults (See Figure 2 and Table 1). Linear mixed modeling also suggested that along with gender, neither age nor height was a significant predictor in PM_2.5 exposures among children (see table S5 for details).

To examine why our study found that children had the highest exposures, we stratified the personal PM_2.5 exposure concentrations by hour of the day (Figure 3).

Figure 3.

Boxplots display the diurnal variation in PM_2.5 concentrations for personal exposures among men, women, and children. The black dotted line shows the median ambient air PM_2.5 concentration. Hours 0 – 5 have been removed from plot due to similarity to other early morning hours. Diamonds represent arithmetic means of PM_2.5 exposure for each demographic-hour.

Diurnal exposure patterns revealed that participants tend to receive their highest exposures between the hours of 18:00 and 20:00. During the daily peak at 19:00, children’s exposures diverge substantially from men’s and women’s, with children exposed to 47% (CI: 33 – 61%) higher concentrations than women and 100% (CI: 79 – 122%) higher than men when accounting for household as a random effect (See Table S6 for model details). This diurnal peak (which coincided with observed dinner times) suggests that the majority of personal PM_2.5 exposures for this population occur during evening meal preparation. These data suggest that children likely participate in cooking or fire maintenance tasks and may also be near the fire as a source of lighting to complete homework during this time window.⁵⁰ These notions were supported by observational reports from project field staff.

Spatiotemporal Exposure Analysis

Our spatiotemporal exposure apportionment assigned 91.4% of all recorded observations to a classified microenvironment, with the remaining 8.6% labeled as ‘no GPS’ observations. Men, women, and children had an equal distribution of time apportioned to microenvironments, 90.0, 92.0, and 91.7%, respectively. Hourly microenvironmental apportionment results are depicted in Figure 4, stratified by demographic group. This figure indicates the average percentage of time that our participants spent in each microenvironment, for each hour of the day.

Figure 4.

(A) Spatiotemporal distribution plot showing average percent of time spent in various microenvironments at each hour of the day among the different demographic groups. (B) The microenvironmental make up of full 48-hour sample averages among demographics. These data are only representative of weekday activity patterns, as we did not sample participants on the weekend.

During the peak PM_2.5 exposure hour (19:00), children were at home 67% of the time, women 63%, and men 50%. Men spent a lot of time in “other” environments throughout the day with some staying in these locations until 20:00, after which most returned home. Children did not spend as much time working in the fields compared to men and women. Instead, children were more likely to be in school during the middle of the day, typically returning home around 17:00. Adults, men in particular, traveled outside the village during morning hours, with the average microenvironmental apportionment for ‘In Transit (Outside Village)’ peaking around 8:00. Similarly, the average ‘Field’ microenvironment apportionment peaked in the morning for both men and women, around 8:00. These activity patterns suggest adults often left the home in the morning to run errands and do agricultural work before returning home, or for some men, to ‘other’ environments around mid-day.

Shown in Figure 5 are average diurnal distributions of PM_2.5 exposure by microenvironment, depicting the timing, location, and magnitude of PM_2.5 exposures for the three demographic groups. In this figure, overall bar heights represent hourly average exposure across participant groups, with the proportion of exposure by microenvironment indicated by color.

Figure 5.

(A) Spatiotemporal distribution plot showing, on average, when and where different demographics received proportions of their daily PM_2.5 exposures. (B) The full 48-hour sample average PM_2.5 exposure variation between demographics among microenvironments. These data are only representative of weekday activity patterns, as we did not sample participants on the weekend.

Participants received the majority of their daily PM_2.5 exposure at home, suggesting that localized cookstove emissions within the home were the largest contributing source for this population’s PM_2.5 exposures (see Figure S14 for a zoomed in version highlighting non-home environments). Adult women spent an average of 58% of their day at home, compared with men at 50%, and children at 53%. Women were also the only demographic likely to receive a large portion of their PM_2.5 exposure around 11:00, suggesting some women performed a mid-day cooking event independent of men and children, who were most likely at school or in “other” environments at this time. However, when children were home in the evenings, they experienced the highest overall increase in PM_2.5 exposure among all the demographics, with exposures 47% (CI: 33 – 61%) higher than the woman and 100% (CI: 79 – 122%) higher than the man in their home at 19:00. Anecdotal evidence from our field staff suggests that children are often tasked with starting and maintaining cooking fires when they are home and available, while the women do most of the food preparation and cooking. Children are also known to use fire for lighting in the evenings while completing their homework.⁵⁰ These behaviors are a possible explanation for why children often have the highest in-home exposures, since the ignition and refueling steps are often the highest polluting stages of working with fire.^51,52

To corroborate observations from our field staff suggesting that children perform most fire maintenance tasks, we examine in Figure S15 how adult women’s PM_2.5 exposures changed at different times of the day, depending on whether their child was present. Although we only monitored one child per household (and other children who did not participate in the study were unaccounted for in our exposure assessment), the results in Figure S15 suggest that women’s exposure to PM_2.5 was reduced when this child was present at home during the 19:00 hour.

For most of the day, women’s in-home exposures were unaffected by whether they had the child being sampled home with them or not. However, at 19:00, women who were home with the monitored child had exposures 4% (CI: 2 – 7%) lower than women who were ‘alone’ (see Table S7 for details). This provides additional evidence to suggest that many children do help with fire maintenance tasks, because having a child home with the woman in the evening on average significantly lowered her PM_2.5 exposures. While this reported 4% difference is small, on average households in the study had 2.3 children between the ages of 5 and 17, of which only one was monitored. Therefore, it is likely that the woman’s decrease in exposure would be even more significant if all the children were accounted for rather than just the one monitored child.

Within the home environment, men were exposed to 48.9% less PM_2.5 than women and 54.7% less than children (see Table S8), on average. This suggests a behavioral difference between men, women, and children within the home environment. Observational evidence from the field team alludes to men performing almost no meal preparation work compared to women and children, with quantitative evidence from our spatiotemporal analysis supporting this notion (Figure S16). This analysis suggests clean cooking access may greatly reduce the PM exposure burden for women and children within this and similar rural communities throughout the world.

Seasonal cycles also played a role in determining individual’s PM_2.5 exposures, with the dry seasons (June to August and December to April) on average leading to 30% (CI: 19 – 43%) higher personal exposure concentrations than the wet seasons (see Table S9 for details). Weather data collected by a nearby Rwanda meteorological weather station (NDEGO AWS6 Station, about 2 km from our study site) suggests the study region received 7.9 times more rainfall during the wet season vs. the dry, scrubbing much of the ambient PM_2.5 from the atmosphere, increasing the humidity in the region, and decreasing the background PM_2.5 pollution levels. The seasonal difference in personal exposure is likely due to several factors, such as dust suppression (from elevated humidity) and washout of PM_2.5 from rainfall during the wet season. Similar seasonal variations in ambient PM_2.5 levels have been reported previously for sub-Saharan Africa.^53–55

Figure S17 shows our reported differences in sample average PM_2.5 exposures among the demographics and between the seasons. We should also note that children had significantly lower exposures during schooling months than during their school-break months (July and August), with exposures 52% (CI: 44 – 60%) lower during their schooling months (see Table S10 for details). We want to clarify that seasonal effects do not impact our household-level analyses, since each family member from a household was sampled once during the same 48-hr period.

Supplementary Material

Supporting Information

Technical and more detailed information on study description and analysis techniques

Short Synopsis Statement.

Spatiotemporal analyses reveal differences in personal PM_2.5 exposures among adults (men, women) and children (aged 8–17) living in the same household within a rural Rwandan community.