A Bayesian Spatio–Temporal Model for Estimating Daily Nitrogen Dioxide Levels

Abstract

A Bayesian spatio-temporal model for estimating daily nitrogen dioxide (NO₂) levels is described. The model uses two datasets with different temporal resolutions. The dataset from the Study of Traffic, Air quality and Respiratory health in children (STAR) contains NO₂ measurements at a relatively large number of sites (most of which are in the state of Connecticut) but for one month in each season over a year. The dataset from the U.S. Environmental Protection Agency (EPA) contains measurements on an hourly level but only at a limited number of sites (four sites in Connecticut). The modal first establishes relationship between STAR observations and EPA observations on the monthly level (the duration-level of the STAR study). The relationship is then assumed to hold at the daily level and thus daily NO ₂ levels at the STAR study sites can be estimated from the average daily NO₂ levels at the EPA sites. The model can also provide predictions of daily NO₂ levels at random sites. The model performed well (R ² > 0.7) and two important implications follow. First, daily pollution information as estimated by the model makes it possible to study the relationship between pollution and pollution-related symptoms such as childhood asthma severity in a more meaningful way. Second, the model offers significant cost reduction on future studies of pollution levels: the readily available pollutant information at EPA monitoring sites and the observations at some sites can be used to make predictions over a long period and at random sites. The model is implemented under the Bayes framework with a Gibbs sampler.