Bayesian Spatio-Temporal Modelling to Deliver More Accurate and Instantaneous Air Pollution Forecasts

Abstract

Air pollution is known to have a significant health impact particularly on people suffering from asthma and other forms of respiratory diseases. In the US ozone pollution is a huge concern during summer months because strong sunlight and hot weather result in harmful ozone concentrations in the atmosphere. Many urban and suburban areas have high levels of ozone concentrations, but many rural areas also have high ozone levels as winds carry emissions hundreds of miles from their sources. With air quality changing day to day, and even hour to hour, the challenge is to devise a model that could provide more accurate forecasts in real time. A Bayesian hierarchical space–time model is proposed and is validated to be the most accurate one that reduces forecasting error up to a third. The method combines observational air monitoring data with a forecast numerical model output to create a statistical model that could be used to provide very accurate forecast maps for the current eight-hour average and the next day maximum eight-hour average ozone concentration levels. The method is fully Bayesian and is able to instantly update the 8-h map at the current hour (upon receiving monitor data for the current hour) and forecast the map for several hours ahead. Consequently, children and vulnerable people suffering from respiratory illnesses could gain potential health benefits by limiting their exposure to potentially harmful air pollution by reducing their outdoor activity when levels are high.