Predicting health impacts of wildfire smoke in Amazonas basin, Brazil.

Worldwide, smoke from forest fires has deleterious health effects. Even so, because of the complexity of fire mechanics, public health authorities face challenges in forecasting and thus mitigating population exposure to smoke. The population in the Amazon basin regularly suffers from fire smoke tied to agriculture and land-use change. The people of Manaus, a city of two million in the center of the basin, suffer the consequences. The study herein evaluates the time lag between fire occurrence and hospital admission for cardiorespiratory illness. Understanding the time lag is key to forecasting and mitigating the public health effects. The study approach is sequential application of four increasingly complex methods of machine learning to examine the relationships among black carbon concentrations, fire count, meteorology, and hospital admissions. The mean absolute percentage error (MAPE) for predicting hospital admissions ranged from 27% to 38%. Furthermore, a one-day lag was observed between the detection of fires and the manifestations of respiratory health hazards. This finding suggests the potential for developing an early warning system, which could enable public health officials to issue advisories or implement preventive actions during the brief period before hospital admissions begin to rise. The findings have applicability not only to the population exposed to fires in the Amazon basin but also to populations where smoke is prevalent, notably increasingly in Australia, southern Europe, the western USA, southern Canada, and southeast Asia.