A critical revision of the peak over threshold approach for flood frequency analysis

In flood frequency modelling the peak over threshold (POT) procedure is considered a valid alternative to the usual analysis of annual maximum serie (AMS). The basic idea is to increase the available information by using more than one flood peak per year. However, in the practical applications ambiguous criteria for peak selection affect the efficiency of the method. An objective procedure for peak identification is proposed here, with which some technical aspects that accompany the application of the POT can be reconsidered. In previous applications the average annual number of events Λ was limited to 2-4 in order to preserve peaks mutual independence. With the revised procedure it is found that larger Λ values, justified by physical and statistical reasoning, can be selected preserving the model basic hypotheses. The problems deriving from the interaction with seasonal runoff components and possible non-stationarities of the time series are also addressed by means of an evolution of the above peak selection method. The proposed procedure is applied to more than 30 time series of daily discharge, showing significant advantages towards the classical POT method.