Bayesian spatio-temporal modelling for forecasting ground level ozone concentration levels

Accurate, instantaneous and high resolution spatial air-quality information can better inform the public and regulatory agencies of the air pollution levels that could cause adverse health effects. The most direct way to obtain accurate air quality information is from measurements made at surface monitoring stations across a study region of interest. Typically, however, air monitoring sites are sparsely and irregularly spaced over large areas. That is why, it is now very important to develop space-time models for air pollution which can produce accurate spatial predictions and temporal forecasts.<br/><br/>This thesis focuses on developing spatio-temporal models for interpolating and forecasting ground level ozone concentration levels over a vast study region in the eastern United States. These models incorporate output from a computer simulation model known as the Community Multi-scale Air Quality (Eta-CMAQ) forecast model that can forecast up to 24 hours in advance. However, these forecasts are known to be biased. The models proposed here<br/>are shown to improve upon these forecasts for a two-week study period during August 2005.<br/><br/>The forecasting problems in both hourly and daily time units are investigated in detail. A fast method, based on Gaussian models is constructed for instantaneous interpolation and forecasts of hourly data. A more complex<br/>dynamic model, requiring the use of Markov chain Monte Carlo (MCMC) techniques, is developed for forecasting daily ozone concentration levels. A set of model validation analyses shows that the prediction maps that are generated by the aforementioned models are more accurate than the maps based solely on the Eta-CMAQ forecast data. A non-Gaussian measurement error model is also considered when forecasting the extreme levels of ozone concentration. All of the methods presented are based on Bayesian methods and MCMC sampling techniques are used in exploring posterior and predictive distributions.

Name dc:type.qualificationname

Ph.D.

Level dc:type.qualificationlevel

doctoral

Grantor dc:publisher.institution

University of Southampton

Year dc:date.issued

2010

Author dc:creator

• Yip, Chun Yin

Advisors dc:contributor.advisor

• Sahu, S.K.
• Forster, J.J.