Bayesian Extreme Value Analysis of Extreme Sea Levels along the German Baltic Coast Using Historical Information

Developed coastlines require considerable investments into coastal protection measures to mitigate the effects of flooding caused by extreme sea levels (ESLs). To maximize the effectiveness of these measures, accurate estimates of the underlying hazard are needed. These estimates are typically determined by performing extreme value analysis on a sample of events taken from tide-gauge observations. However, such records are often limited in duration, and the resulting estimates may be highly uncertain. Furthermore, short records make it difficult to assess whether exceptionally large events within the record are appropriate for analysis or should be disregarded as outliers. In this study, we explore how historical information can be used to address both of these issues for the case of the German Baltic coast. We apply a Bayesian Markov chain Monte Carlo approach to assess ESLs using both systematic tide-gauge observations and historical information at seven locations. Apart from the benefits provided by incorporating historical information in extreme value analysis, which include reduced estimate uncertainties and the reclassification of outliers into useful samples, we find that the current tide-gauge records in the region alone are insufficient for providing accurate estimates of ESLs for the planning of coastal protection. We find long-range dependence in the series of ESLs at the site of Travemünde, which suggests the presence of some long-term variability affecting events in the region. We show that ESL activity over the full period of systematic observation has been relatively low. Consequently, analyses which consider only these data are prone to underestimations.