Knowledge about extreme sea-levels is essential for prediction of flooding risks, coastal management and design of coastal infrastructure systems. Furthermore, it assumes particular relevance in a climate change context: long-term changes in extreme sea-levels may be associated with regional climatic variability, and coastal regions may face an increased flooding risk, particularly for areas where the combination of local subsidence and global sea-level rise enhances the rate of relative sea-level change. In studies of regional sea-level variability, tide gauge records are often analyzed individually for characterizing sea-level variability at each location. Marginal analysis, however, is in itself insufficient to come to an accurate description of regional sea-level variability. An alternative approach is to consider simultaneously the whole data set of sea-level records from a given region, and characterize regional variability in terms of locations exhibiting similar behaviour, through clustering techniques.
The purpose of this talk is two-fold: first, a new time series clustering approach which combines Bayesian methodology, extreme value theory and classification techniques is presented for the analysis of the regional variability of sea-level extremes. The tide gauge records are clustered on the basis of their corresponding predictive distributions for 25-, 50- and 100-years return values. Second, a clustering method for time series based on short-term predictions of forecast densities is also presented in order to describe regional sea-level variability in the Baltic Sea.