Household and community poverty, biomass use, and air pollution in Accra, Ghana

NM has the highest population density (441 people per 10,000 m²), followed by JT (329 per 10,000 m²), AD (27 per 10,000 m²), and EL (5 per 10,000 m²) (Fig. 1A). The SES index in census enumeration areas (EAs) in JT and NM are in the lowest quintile of all EAs. In contrast, the SES of AD and EL fall into the wealthiest quintile (Fig. 1B). In the census, about 80% of households in JT and NM used biomass fuels, compared with 43% in AD and 53% in EL (Fig. 1C). In our study households, biomass use was highest in JT, where 95% [95% confidence interval (CI) 85–100%] and 45% (23–67%) of households used biomass for their own and small-commercial cooking, respectively (Table S1). At the low end, only 22% (3–41%) and 6% (0–17%) of surveyed households in EL used biomass for their own and small-commercial cooking. EL was surrounded by other high-SES and below-median biomass communities. The other three neighborhoods were closer to the city center and were surrounded by communities that may have had lower or higher SES and biomass use prevalence (Fig. 1).

The housing arrangement in most study households in JT (90%) and NM (100%) was a compound room, with multiple households living in different parts of a larger single structure built around a central courtyard. Most households in JT and NM cooked outdoors in the open-air shared compound courtyard, where their neighbors may also cook (Table S1). In EL, 89% of study households lived in separate, free-standing houses, and 83% of the households cooked in separate indoor kitchens (Table S1).

In most households in AD and JT, cooking area PM_2.5 (particles with aerodynamic diameter ≤ 2.5 μm; fine PM) concentrations were lower than ambient levels (Fig. 2A), with geometric mean household-to-ambient ratios of about 0.70. Some cooking area PM_2.5 concentrations in NM were lower than the ambient levels, whereas others were higher, leading to a geometric mean household-to-ambient ratio of 0.97. Cooking area PM_2.5 was similar to the ambient levels in most EL households. However, the household-to-ambient ratios had a geometric mean of 1.22 because of higher cooking area concentrations in five households. Two of these five households used charcoal as their primary fuel, another raised poultry on their compound.

Unlike PM_2.5, cooking area coarse PM (PM_2.5–10) concentrations exceeded corresponding ambient levels everywhere, except in one AD household (Fig. 2B). Mean residual cooking area PM_2.5–10 (cooking area minus ambient) was 58 μg/m³ and mean household-to-ambient ratio was 2.32. These results suggest the disproportionate presence of household sources for coarse PM, such as sweeping and resuspension. As a result of such fine and coarse PM patterns, cooking area PM_2.5-to-PM₁₀ ratios were lower than the ambient ratios on the same day (Fig. 2C).

Cooking area PM was lowest in EL, with geometric means of 25 (21–29) μg/m³ for PM_2.5 and 28 (23–33) μg/m³ for PM_2.5–10, and highest in JT with 71 (62–80) μg/m³ for PM_2.5 and 118 (101–138) μg/m³ for PM_2.5–10, and in NM with 52 (44–63) μg/m³ for PM_2.5 and 131 (114–150) μg/m³ for PM_2.5–10. Although measurement periods varied, data on seasonal patterns of ambient PM reported elsewhere (2) and the above household-ambient comparisons suggest that PM in JT households would likely be the highest of all neighborhoods, regardless of season, as was ambient PM in this neighborhood.

Using biomass fuels and living in a high biomass-use community were both associated with higher cooking area PM (high vs. low biomass-use communities were defined based on whether the proportion of households using biomass fuels in the EA was above vs. below median of all EAs) (Tables 1 and 2). The lowest cooking area PM_2.5 and PM_2.5–10 were measured in homes with clean fuels and in low biomass-use communities, 27 (24–31) and 45 (34–59) μg/m³, respectively, and the highest in homes that used biomass and were in high biomass-use communities, 60 (53–68) and 128 (116–142) μg/m³, respectively. Of the other two groups, cooking areas in high biomass-use communities with clean fuels had higher PM than those that used biomass fuels but lived in low biomass-use communities. Similarly, commercial cooking with biomass fuels was associated with higher cooking area PM, making such households when located in high biomass-use communities the most polluted, with PM_2.5 and PM_2.5–10 geometric means of 77 (64–91) and 143 (120–169) μg/m³, respectively (Table 2). There was no meaningful difference between homes that did no commercial cooking and those that did so with clean fuels, but the sample size for the latter was four and hence should be considered as suggestive only.

The results in Tables 1 and 2 show crude associations, without controlling for other variables that may vary across households. The multivariate associations confirm that using biomass fuels for own and commercial cooking, and living in EAs with higher biomass-use prevalence, were associated with higher cooking area PM; only the effects of neighborhood fuel use were consistently significant (Table 3). Beyond their statistical significance, living in an EA with 26% higher biomass-use prevalence had about the same effect on cooking area PM_2.5 as switching from a cleaner fuel to biomass in model 1; the equivalence would be at 69% higher biomass-use prevalence for PM_2.5–10. A household located in an EA where all households use biomass fuels would have 149% (104–223%) of the PM_2.5 level and 165% (122–246%) of the PM_2.5–10 level compared with its counterpart in an EA with no biomass use after adjustment for neighborhood ambient PM; the effects were 272% (182–406%) and 272% (165–448%) in the model that did not adjust for ambient PM. For fine PM, the (proportional) effects of using biomass for commercial cooking seemed larger than using it for household purposes, whereas the opposite was seen for coarse PM, although the differences were not statistically significant. The associations with household size, average distance to main roads, cooking location, and the presence of smokers in the house were generally nonsignificant. Adjusting for neighborhood PM weakened the association with EA biomass use for both size fractions, and that of household biomass use for PM_2.5, but the effect of household biomass use on PM_2.5–10 became larger and significant after this adjustment.

In all neighborhoods, both the ambient and cooking area PM_2.5 rose in the early morning hours. This morning rise started as early as 0300 hours in JT and NM vs. around 0600 hours in EL (Fig. 3). Although in any single neighborhood this pattern may either be because of morning residential and small commercial cooking, other commercial activities that use biomass (e.g., fish smoking and bakeries), and traffic, or because of overnight surface temperature inversions, the differences in start time and rise across neighborhoods make the differential patterns of sources a more likely explanation. Specifically, in both JT and NM, cooking street food and other activities that use biomass fuels begin at very early hours. In JT and NM, we also observed a midday peak around 1100 hours, which may correspond to midday cooking and traffic. As described elsewhere (2), ambient PM also showed a rise in PM_2.5 in the evening (1800–2100 hours) except in EL, possibly because of evening rush hour and biomass use; this evening rise was less noticeable in cooking areas. In JT and AD, ambient PM_2.5 was higher than cooking-area levels, whereas the two environments had similar PM_2.5 in EL and NM.

On average, PM_2.5 concentrations in the cooking and living areas tracked relatively well, suggesting diffusion between household environments or from the ambient air to household environments (Fig. 3). However, the pairwise correlations between continuous PM_2.5 in different indoor and ambient environments varied substantially, with living area concentration in households using nonbiomass fuels having higher correlation with ambient levels than those that used biomass (Fig. S2).