Evaluation of design flood estimates - a case study for Norway

Abstract

The aim of this study was to evaluate the predictive fit of probability distributions to annual maximum flood data, and in particular to evaluate (i) which combination of distribution and estimation method gives the best fit and (ii) whether the answer to (i) depends on record length. These aims were achieved by assessing the sensitivity to record length of the predictive performance of several probability distributions. A bootstrapping approach was used by resampling (with replacement) record lengths of 30 to 90 years (50 resamples for each record length) from the original record and fitting distributions to these subsamples. Subsequently, the fits were evaluated according to several goodness-of-fit measures and to the variability of the predicted flood quantiles. Our initial hypothesis that shorter records favor two-parameter distributions was not clearly supported. The ordinary moments method was the most stable while providing equivalent goodness-of-fit.