A spatial covariates model for storm surge extremes in the German Bight

Storm surges and accompanying extreme water levels pose a major threat to coastal structures, urban and industrial areas and human life in general. In order to develop effective risk mitigation strategies, it is crucial to improve the understanding of these extreme events as well as their occurrence probabilities and quantiles, respectively.

The standard procedure to estimate extreme quantiles (return-levels) is to fit a suitable distribution to the observed extreme values on a site-by-site basis. However, this approach exhibits some disadvantages: 1) Estimates of extreme quantiles are only available at gauged locations. 2) The small amount of extreme events in tide gauge records makes these estimates highly uncertain.

We tackle both issues by pooling all available tide gauge records together through a covariates model that allows for smoothly varying distribution parameters in space. Using this approach, the model is not only able to reduce the uncertainty in quantile estimates, but also enables the interpolation of the distribution parameters at arbitrary ungauged locations, e.g. in between tide gauge locations.

Deploying our model for the German North Sea coast, we generate a probabilistic reanalysis of extreme water levels as well as associated probabilities for the period 2000 – 2019.