Winter, Hugo and Tawn, Jonathan (2016) Extreme value modelling of heatwaves. PhD thesis, Lancaster University.
2016winterphd.pdf - Published Version
Available under License Creative Commons Attribution-NonCommercial-NoDerivs.
Download (3MB)
Abstract
Since the turn of the century record temperatures have been observed in at least 20 different countries across Europe. Isolated hot days are not often an issue; most devastation occurs when hot temperatures persist over many days. For example, the 2003 heatwave over Europe caused 40,000 deaths over a four week period at a cost of e 13.1 million to the agriculture sector. It is clear that accurate models for the risks associated with heatwaves are important to decision makers and planners who wish to reduce the number of people affected by these extreme events. Extreme value theory provides a statistical framework for modelling extreme events. Extreme value models for temperature data tend to focus solely on the intensity, overlooking how long periods of hot weather will last and what the spatial extent of the event will be. For heatwaves, it is vital to explicitly model extremal dependence in time and space. An aim of this thesis is to develop extreme value methods that can accurately capture the temporal evolution of heatwaves. Specifically, this is the first to use a broad class of asymptotically motivated dependence structures that can provide accurate inferences for different types of extremal dependence and over different orders of lagged dependence. This flexibility ensures that these models are less likely to dramatically under or over-estimate the risks of heatwave events. Climate change is now widely regarded as a driving force behind increased global temperatures. Extending the extreme value heatwave models to include covariate structure permits answers to critical questions such as: How will a 1°C warming in the global temperature increase the chance of a 2003 style event? The 2009 heatwave over Australia highlighted issues posed when multiple cities are affected simultaneously. Both Adelaide and Melbourne observed record temperatures during the same event which led to 374 deaths and 2000 people being treated for heat related illness. It is not enough for heatwave models to account for temporal dependence, they also need to explicitly model spatial dependence. Large-scale climatic phenomena such as the El Nino-Southern Oscillation are known to affect the temperatures across Australia. This thesis develops new spatial extreme value methods that account for covariates, which are shown to model the 2009 event well. A novel suite of spatial and temporal risk measures is designed to better understand whether these covariates have an effect on the spatial extent and duration of heatwaves. This provides important information for decision makers that is not available using current methodology.