Nonparametric variable kernel estimation with historical floods and paleoflood information

Abstract

Extending a data record back in time using historical or paleoflood data has the potential to provide a considerable amount of information on very large floods. Parametric estimation methods are readily applicable to flood frequency analysis when historical or paleoflood estimates are available. However, all parametric approaches need an assumption about the underlying parent distribution, which is never known exactly. In recent years nonparametric methods of estimating probability density functions have been developed. Each of these involves the use of a suitable smoothing function known as a kernel. A new nonparametric variable kernel estimation model is proposed. Results obtained from a limited amount of real data and from simulation experiments show that quantiles estimated by the nonparametric method are better than those estimated by some selected parametric models both in descriptive ability and in predictive ability. The uncertainty in the choice of the threshold value of perception of a historical flood is also discussed.