Ensemble approach for projections of return periods of extreme water levels in Estonian waters

Abstract

The contribution of various drivers to the water level in the eastern Baltic Sea and the presence of outliers in the time series of observed and hindcast water level lead to large spreading of projections of future extreme water levels. We explore the options for using an ensemble of projections to more reliably evaluate return periods of extreme water levels. An example of such an ensemble is constructed by means of fitting several sets of block maxima (annual maxima and stormy season maxima) with a Generalised Extreme Value, Gumbel and Weibull distribution. The ensemble involves projections based on two data sets (resolution of 6h and 1h) hindcast by the Rossby Centre Ocean model (RCO; Swedish Meteorological and Hydrological Institute) and observed data from four representative sites along the Estonian coast. The observed data are transferred into the grid cells of the RCO model using the HIROMB model and a linear regression. For coastal segments where the observations represent the offshore water level well, the overall appearance of the ensembles signals that the errors of single projections are randomly distributed and that the median of the ensemble provides a sensible projection. For locations where the observed water level involves local effects (e.g. wave set-up) the block maxima are split into clearly separated populations. The resulting ensemble consists of two distinct clusters, the difference between which can be interpreted as a measure of the impact of local features on the water level observations.