Repeated assessment of PM2.5 in Guatemalan kitchens cooking with wood: Implications for measurement strategies

Household air pollution resulting from solid fuel combustion is a leading cause of global morbidity and mortality. Strategies to measure area concentrations of and exposures to PM2.5 in rural homes focus primarily on short-term measurements, often of 24 or 48 h. Little is known about how well these short-term measurements, commonly used exposure metrics in health risk assessment of the impacts of household air pollution exposure, predict longer-term averages. In San Lorenzo District, Guatemala, we deployed the relatively low-cost University of California, Berkeley (UCB) Particle and Temperature Sensor (PATS) for 120-333 days in the kitchens of 8 homes using biomass fuels. We evaluated how well short-term measurements predicted the household-level, entire -sample average. A single 24-h measurement had between a 32% and 39% chance of being within +/- 25% of the household-level mean of all measurements. The Root Mean Square Error (RMSE) of a single 24-h measurement was on average 4.5 times higher than that of the mean of measurements taken once per study week. Alternate strategies - including sampling once per study week or once per study month - with this class of lower-cost sensors yield estimates which have a higher probability of being closer to the overall average value and have smaller errors relative to the overall mean. Evaluation of how well short-term measures predict longer-term averages of household air pollution at prospective study sites allows optimization of field resources to better estimate indoor concentrations and personal exposures.

Automated summary

Household air pollution from solid fuel combustion is a major cause of global morbidity and mortality. Short-term measurements are commonly used to assess the health risks of household air pollution exposure, but their ability to predict longer-term averages is not well understood. In Guatemala, we used a low-cost sensor to measure particle and temperature levels in the kitchens of 8 homes over 120-333 days. We found that a single 24-hour measurement had a 32-39% chance of being within 25% of the household-level mean, and the RMSE of a single 24-hour measurement was 4.5 times higher than the mean of weekly measurements. Sampling once per week or per month yielded estimates that were closer to the overall average and had smaller errors.