Effects of disregarding seasonality on the distribution of hydrological extremes

Abstract

This paper deals with the seasonality of hydroclimatic extremes and with the problem of accounting for their non-homogeneous character in determining the design value. To this aim we devise a simple stochastic experiment in which extremes are produced by a non-homogeneous extreme value generation process. The design values are estimated in closed analytical form both in a peak over threshold framework and by using the standard annual maxima approach. In this completely controlled world of generated hydrological extremes, a statistical measure of the error associated to the adoption of a homogeneous model is introduced. The sensitivity of this measure, named return period ratio, to the typology and strength of seasonality is investigated. We find that neglecting seasonality induces a downward bias in design value estimators. The magnitude of the bias may be large when the peak over threshold approach is adopted, while the return period distortion is limited when the annual maxima are considered. An application to rainfall data from a 30 000 km² region located in North-Western Italy is presented to better clarify the effects of disregarding seasonality in a real case.