ABSTRACT
Using a frequency model consists in part of using the relation that provides the value of a quantile xq as a function of the cumulative frequency q = F(xq) or, conversely, of computing the cumulative frequency (and thus the return period) corresponding to a given value x. This aspect, which may seem self-evident, was already addressed several times, especially in Chapters 1 and 2, and will be described only briefly here. In addition we can compute the probability of occurrence of a particular hydrological event. These computations generally operate on the assumption that the adopted frequency model refers to annual probabilities or that the observation periods (the statistical trials) are in years. However there can be cases in which the analysis does not deal with annual periods, especially when we are trying to take into account as much of the available data as possible, with the ongoing aim of comprehensiveness and homogeneity. This is the case, for example, when we are performing an analysis of POT series or series of k greatest annual events (“inflated” series). In such instances we are dealing with infra-annual series and it is advisable, before computing the probability of occurrence, to transform the infra-annual series to annual series – an operation we will call annualization. Next we will look at an example of applying some frequency models in order to reach a more general synthesis: IDF curves (intensity - duration - frequency) which summarize in a single model a series of frequency models that each corresponds to a different rainfall duration. We will examine next a method that solves the composition of errors problem by exposing a little known approach: the Rosenblueth method.
