Reed, D.W. 2008b. Inter-site dependence in extremes: unlocking extra information. Proc. FLOODrisk2008 International Conf., Oxford, 30 September - 2 October 2008.
For full paper: Click here
Where long records have been gathered at a relevant site, and are thought to represent a near-stationary process, their statistical analysis is pivotal to environmental risk estimation. In other situations, pooling data from nearby or similar sites provides a pragmatic way of stretching estimates to the long return period typically required. A well-constructed pooling scheme promotes site-to-site consistency in design values, i.e. estimates of T-year extreme values. This fits well with the “we estimate for all sites” mindset now fashionable. The paper explores the concept that the study of inter-site dependence yields valuable extra information. In this context, dependence is the tendency for extremes at neighbouring sites to occur together. Rank correlation in annual maxima provides a simple measure of pairwise dependence in extreme values at different sites. Results are presented from a number of studies of dependence based on flood data taken from HiFlows-UK. These include case examples of the Spey, the Yorkshire Ouse, three rivers in south-east Wales and nine urbanised catchments in the London area. Sites are mapped by multi-dimensional scaling so that inter-site distance reflects the degree of independence in their extremes. The paper concludes that dependence analysis can inform a range of applications.