PM2.5 forecasting in Coyhaique, the most polluted city in the Americas

Abstract

Coyhaique is a southern Chilean city with a population of approximately 64,000 habitants. In spite of its small size, Coyhaique has been identified as the city with highest annual PM2.5 concentrations of the Americas (including south America, central America and north America). Episodes of high pollution are concentrated on the fall- winter season when meteorological conditions do not favor atmospheric particle dispersion and extended use of wood stoves is responsible for more than 99% of the emissions. In Chile, the 24 h average of PM2.5 concentration is classified in four ranges: fair, bad, very bad and critical. We have developed a neural network model and a linear model aimed to forecast the maximum of the 24 h moving average one day in advance. Input variables for the models are hourly values of PM2.5 at 18 h and 19 h of the present day, measured and forecasted temperature, wind speed and precipitation and measured values of NO2, CO and O3 concentrations. The neural network model is slightly more accurate than the linear model. We are able to anticipate the observed range in 75% of the cases, and critical days in 84% of the cases.