A large data set of 240 catchments scattered all around France is used, and in particular, we benefit from a high- resolution precipitation database (ANTILOPE, Météo-France) that describes hourly precipitation at 1 km 2 resolution. Data were aggregated at different time steps (1h, 3h, 6h, 12h and 24h). Streamflow simulations are performed at these different time steps using the GR5 model in its lumped and semi-distributed version (GRSD, de Lavenne et al. (2016)), with a mesh grid of 500, 250, 100 and 50 km 2. Ten different indices are used to describe spatio- temporal characteristics of rainfall events, in order to analyse in which contexts refined resolutions are needed to improve the performance of the model. These indices characterise the spatial variability, localisation, movement, intensity and temporal variability of rainfall events. In addition to some indices already reported in the hydrological literature, we propose some new indices like an indice usually applied in economics. This analysis at different time steps, events and catchments demonstrates the limits for some of them and allows to propose some corrections (Goullet J., 2016).