La prédétermination des crues

Abstract: Predetermination (or statistical prediction) can be defined as the announcement of the physical and statistical characteristics of a future event non-precisely located in time. So, it is quite different from forecasting, whose objective is to give the precise date of occurrence of a specified physical event. Predetermination will then be inseparable of probabilistic concepts such as the probability of occurrence of a given event or, equivalently, of its return period. About floods, one will estimate, for a given river cross-section, whether the probability that the discharge would exceed a given threshold or, symmetrically, the discharge which has a given probability of exceedance. Such estimations, the spirit of which is definitively different of that of PMP/PMF (Probable Maximum Precipitation or Flood), enable a rational approach of socio-economic problems linked to hydraulic works design and land use. The only available data has long been the instrumental discharge measurements, whose series are rather short, some decades in general, seldom reaching a century. More recently the so-called historical data have been used. They are compiled from old documents and supplemented by hydraulic studies and even geomorphological or sedimentological studies. Such data increase considerably the length of extreme events'' time series available. There is still a great confusion about the statistical models to be used. The choice of such models is too often only dictated by the best graphical fit, while this choice is fundamental for data interpretation. Without a wide theoretical agreement, the results are still not so reliable. A very important standardisation effort has been done in the US, where, since the end of the sixties, the use of the Log-Pearson III distribution has been made compulsory for all federal projects, but the physical and even statistical arguments of this choice remain quite weak. All the efforts done to enlarge the data bases, as well as the theoretical developments based upon the physics of hydrological processes and especially their scale invariance, should nevertheless enable us to formalize in a near future the asymptotical behaviour of hydrological time series and to define effective predetermination methods in engineering. To cite this article: P. Hubert, C. R. Geoscience 337 (2005). [Copyright &y& Elsevier]