Quick flood forecasting with lumped hydrological models. Application on operational upper Loire river catchments: modelling assessment, taking into account uncertainties on mean areal precipitation and use of meteorological forecasts.

The aim of the present work is the evaluation of lumped rainfall-runoff (RR) models for flood forecasting in the case of upper Loire river catchments.<br /><br />Following the description of Loire catchment at Bas-en-Basset, an analysis demonstrates both the worth and the flaws of presently available data sets. The high variability of these hydrometeorological events enables us to compare RR models in particularly difficult but interesting contexts.<br /><br />Simple conceptual models appear more robust and often more efficient than data-driven models. A further evaluation, based on specific criteria for flood forecasting, highlight the information about flood evolution provided by conceptual RR models, even though modelling errors remain altogether significant while the various models behave in a similar way.<br /><br />Estimation of mean areal precipitation is conducted with kriging tools and a model of uncertainty on mean areal precipitation estimation is proposed and validated on data.<br /><br />These uncertainties are then propagated within RR models. Their impact is reasonably different with respect to catchment size, as a variable part of the global modelling error may be explained. <br /><br />Finally, an exploratory work has demonstrated the usefulness of taking into account probabilistic precipitation forecast into a hydrometeorological chain: longer anticipation has consequently been obtained. Although pre-processing linked to these forecast availability is absolutely necessary.<br /><br />As a conclusion, simple tools let us expect improvements in this very perfectible field.