Spatial dependence of river flooding and extreme rainfall

The extent of spatial dependence of flooding and extreme rainfall in Britain is unknown. In this thesis we develop existing extreme value theory to study this phenomenon. It is important to know the level of spatial dependence of flooding because if a flood occurs at multiple sites at the same time the consequences for (for example) the emergency services and the insurance industry can be more severe than if these floods occurred at different times. There are various features of river flow and rainfall data that complicate the estimation of extremal dependence. The two main features are the presence of missing data and the temporal dependence of the individual time series. Heffernan and Tawn (2004) developed a flexible method for modelling multivariate extreme values that can be used for a large number of variables. In this thesis we develop and extend this method to handle the presence of missing data and temporal dependence. We apply this extended method separately to river flow and rainfall data to estimate the level of spatial dependence of flooding and extreme rainfall over Britain. We find that heterogeneous regions in terms of the geology and soil types show low levels of spatial dependence. As we would expect we also find that connected rivers show a high level of dependence between them. We also find that the upland regions of Britain have low levels of spatial dependence of extreme rainfall.