Linking meteorology, air pollution, and health in Melbourne, Australia

2017-01-31T04:46:20Z (GMT) by Pearce, John Lanier
The importance of meteorology in air pollution processes and its influence on human health is understood; however, these relationships and their interactions are expected to be different under a changing climate. Thus, a considerable challenge is presented to those charged with air quality management because alterations in meteorological conditions stemming from a changing climate will vary from one geographic region to the next. To further complicate matters, the influence of weather on air pollution and related health effects also varies geographically. Therefore, in order to better understand these consequences for any given region around the globe, regional scale studies are required. Such studies also elucidate the processes and how they may be similar and different between regions. The aim of this thesis is to assess the relationships between meteorology, air pollution and human health for Melbourne, Australia during the years 1999 to 2006 to provide insight into how this may alter under changing climate conditions. This is achieved using a novel crossdisciplinary approach that draws from the fields of atmospheric science, epidemiology, and statistics. In the first part of the study, the influence of synoptic-scale circulation features on daily concentrations of ozone (O3), particulate matter ! 10 μm (PM10), and nitrogen dioxide (NO2) were characterized by using a synoptic climatology developed using self-organizing maps (SOMs) and applied within the framework of a generalized additive model (GAM). Results demonstrated that large-scale circulation features were not a primary driver of local air quality during our study period. Nevertheless, differential effects were found between circulation features with a general trend of anticyclones being associated with significantly poorer air quality. In particular, NO2 and O3 were 20% higher than average when synoptic conditions resulted in a northeast gradient wind over the region. For PM10, maximum increases of up to 20% over normal concentrations occurred when a strong anticyclone was centered directly over the region. The second part of the study again applied the framework of GAM and characterized the relationship between locally observed weather elements and daily pollutant concentrations. These findings demonstrated that local-scale meteorological conditions were a more important driver of air quality than synoptic-scale circulation. The key finding in this analysis was that when daily maximum temperatures exceeded 35 °C; O3, PM10 and NO2 concentrations were 150%, 150% and 120% higher than the average. Other elements such as winds, boundary layer height, and atmospheric moisture were also important; however, their influences were marginal when compared to temperature. The final part of the study, an ecological epidemiological study, examined the statistical relationship between daily mortality and air pollution across different temperatures using case-crossover analysis (CCO) and GAM. Results showed that temperature behaved as an effect modifier in the air pollution-mortality relationship with each pollutant exhibiting stronger effects on mortality as ambient temperatures increased. These findings, expressed as the percentage change in mortality along with their 95% confidence intervals, were strongest when temperatures exceeded 22 °C as mortality increased 2.82% [0.84, 4.85] per 10 ppb increase in O3, 3.14% [1.57, 4.75] per 10 "g increase in PM10, and 5.05% [1.16, 9.10] per 10 ppb increase in NO2. In conclusion, findings from this thesis provide a direct link between weather, air quality, and human health. Moreover, the strength of weather – in particular temperature – as a driver in these relationships suggests that if current climate projections hold true and all else remains the same then air quality will decline. Finally, characterizing these relationships through the use of statistical analyses is of added benefit as results are a direct reflection of patterns in observed data and thus lack the bias present in deterministic modeling studies.