Extreme hydro-meteorological events and their probabilities

Extreme hydro-meteorological events usually have a large impact on our society. For safety standards regarding life and property and for design purposes of large structures extreme events with return periods between 100 and 10,000 years are often required. A practical difficulty in determining such rare events is due to the fact that our instrumental meteorological records are typically not longer than about 100 years. We are thus interested in extreme events that may never have occurred in the instrumental history. Such extremes are therefore usually estimated by extrapolating a fitted probability distribution. The results obtained with statistical extrapolation methods, however, strongly depend on the assumed probability distribution. An attractive alternative to these classical methods is resampling of historical meteorological time series. Resampling is attractive since it is a nonparametric technique, which means that no assumptions about the underlying distributions of the data have to be made. In addition, resampling offers the opportunity to simulate different meteorological variables (multivariate) for different locations (multi-site) simultaneously, while the cross-correlations (between variables) and the spatial correlations (between locations) are automatically preserved. Resampling, finally, makes it possible to simulate much longer time series than the historical records from which is resampled. Such very long time series usually contain many unprecedented extreme events which can serve in a frequency analysis of the extremes. In short, resampling is a very suitable nonparametric technique to simulate multi-site multivariate meteorological time series that are much longer than those from the instrumental records. With specific hydrological applications in mind such very long resampled time series are used to determine the size and probabilities of occurrence of extremely wet periods in the Rhine basin (that may result in river flooding) and of extreme droughts in the Netherlands (leading to economic losses in agriculture and shipping). Resampling techniques are further used to determine the statistical uncertainty of extreme hydro-meteorological events and of other properties of (hydro-)meteorological data.