Comparison of precipitation extremes estimation using parametric and nonparametric methods

Due to recent occurrence of extreme hydrological events in Central Europe, there is an increasing interest in more accurate prediction of return levels of such events. The precipitation records from 6 ombrographic stations operated by the Czech Hydrometeorological Institute were analyzed in order to estimate the intensity-duration-frequency. Although the longest rainfall series consists of more than 40 years of measurements, data set contains also records from newly established stations with only short-time series available. Impact of the series length on the estimation quality is part of this study. Parametric and nonparametric approaches to drawing samples are assumed. In the first case, we consider a threshold model and we estimate the unknown parameters using maximum likelihood and probability weighted moments methods. In the latter case, k largest order statistics are considered and the bootstrap methodology is applied as a resampling technique together with the moment estimator of extreme value index.