Vers une amélioration d'un modèle global pluie-débit au travers d'une approche comparative

The study examined and evaluated 38 structures of lumped rainfall-runoff models on 429 catchments located in France, the United-States, Australia, the Ivory Coast and Brazil. All structures were evaluated at a daily time-step in the same testing framework and fed with the same amount of data. Their parameters were calibrated using a local optimisation search procedure that proved able to locate satisfactory sets of optimum parameters. Model performances in terms of quality of observed streamflow simulation were assessed with several quantitative criteria, including new formulations. Tests results indicate that a low number of free parameters (3 to 5) is sufficient to achieve a satisfactory modelling of the catchment hydrological behaviour, the increasing number of degrees of freedom in the model structure limiting model robustness. Furthermore model structure plays a key role in the quality of simulations. Some mathematical tools used in different structures appear complementary and their association in a single model may improve streamflow simulations. The structure of the simple GR3J model was thus modified into a more satisfactory four-parameter structure, with improved low flow simulations. Such a model can be used to solve many issues of water resources management and engineering.