Abstract
Event-based models are often used for flood prediction because they require fewer data than morecomplex models and account for a small number of parameters. We present the performance of such amodel in simulating Mediterranean floods, with a focus on the initialization and on the impact of therainfall uncertainties on the calibration of the model. The distributed event-based parsimonious SoilConservation Service Lag-and-Route (SCS-LR) model was applied in the Real Collobrier catchment,France, which has a very high density of raingauges. The initial condition of the model was highlycorrelated with predictors, such as baseflow or soil water content. A reduction in the raingauge densitycan markedly change the calibration of the model. As the density of raingauges is generally low in mostcatchments, the uncertainties associated with rainfall measurement are thus expected either to maskthe actual accuracy of the model, or to alter the model parameters