Title: Probabilistic forecasting of an air quality index
Authors: Jooyoung Jeon - University of Bath (United Kingdom)
Xiaochun Meng - University of Sussex (United Kingdom)
James Taylor - University of Oxford (United Kingdom) [presenting]
Abstract: Air pollution has emerged as a major issue affecting human health, with the resulting burden on health systems having economic and political implications. Urban air pollution is believed to be the cause of more than a million premature deaths worldwide each year. Respiratory illnesses are the main health risk factor, but the prevalence of heart disease, stroke and cancer is also increased. Air quality indices are widely-used to summarise the severity of the level of a set of pollutants, with a traffic light signal often used to provide a visual indicator. The index is a convenient measure used by policy makers, but is also used, on a day-by-day basis, by health professionals and the public, especially those with a history of respiratory conditions. Forecasts of the index are typically produced each day for lead times up to several days ahead. The predictions are usually provided by meteorologists using atmospheric models that have chemistry features incorporated. Probabilistic forecasting from such models is not straightforward, and hence is very rare. We consider the use of time series models to produce density forecasts for the index. The approach involves the fitting of a multivariate model to a set of six pollutants. To capture the dependencies between the pollutants in a practical way, we use an empirical copula. The empirical work uses hourly data from South Korea, where more than half the population are considered to be exposed to dangerous levels of pollutants.