Abstract
Rainfall-runoff models are essential tools for many hydrological applications, including flood forecasting. The purpose of this thesis was to examine the performances of a distributed event model for reproducing the Mediterranean floods. This model reduces the parametrization of the processes to the flood period, and estimates the saturation of the catchment at the beginning of the event with an external predictor, which is easily observable or available. Such predictor avoids modelling the inter-flood phase and simplifies the parametrization and the calibration of the model. The selected model combines a distributed SCS production function and a Lag and Route transfer function, applied to a discretization of the basin in a grid of regular square meshes.The model was first tested on the Real Collobrier watershed. This Mediterranean basin has been monitored by IRSTEA for more than 50 years and has an exceptional density of rainfall and flow measurements. This favourable environment made it possible to reduce the uncertainties on the rainfall input and to evaluate the actual performances of the model. In such conditions, the floods were correctly simulated by using constant parameters for all the events, but the initial condition of the event-based model. This latter was highly correlated to predictors such as the base flow or the soil water content w2 simulated by the SIM model of Meteo-France. The model was then applied by reducing the density of the rain gauges, showing loss of accuracy of the model and biases in the model parameters for lower densities, which are representative of most of the catchments.The spatial variability of the model parameters was then studied in different Real Collobrier sub-basins. The comparison made it possible to highlight and correct the scale effect concerning one of the parameters of the transfer function. The catchment saturation predictors and the initial condition of the model were still highly correlated, but the relationships differed from some sub-catchments. Finally, the spatial variability of the model parameters was studied for other larger Mediterranean catchments, of which area ranged from some tenth to hundreds of square kilometres. Once more, the model could be efficiently initialized by the base flow and the water content w2, but significant differences were found from a catchment to another. Such differences could be explained by uncertainties affecting as well the rainfall estimation as the selected predictors. However, the relationships between the initial condition of the model and the water content w2 were close together for a given type of catchment.In conclusion, this distributed event model represents an excellent compromise between performance and ease of implementation. The performances are satisfactory for a given catchment or a given type of catchment. The transposition of the model to ungauged catchment is less satisfactory, and other catchment saturation indicators need to be tested, e.g. in situ measurements or satellite measurements of soil moisture.