PERFORMANCE OF L-MOMENTS AND NONPARAMETRIC FLOOD FREQUENCY-ANALYSIS

A simulation study was undertaken to compare parametric L-moments and nonparametric approaches in flood frequency analysis. Data of various sample lengths were generated from a given generalized extreme value d istribution and the quantiles estimated using the fixed-kernel nonpara metric method and from a generalized extreme value distribution fitted by L-moments. From the resulting root-mean-square errors for various quantiles, it was concluded for unimodal distributions that nonparamet ric methods are preferable for large return period floods estimated fr om short (<30 years) samples while parametric methods are preferable i n other circumstances. It was also pointed out that nonparametric meth ods are more suitable for mixed distributions.