This thesis aims at exploring the question of temporal scaling in lumped conceptual hydrological modelling. The main objectives of the thesis are to: (i) study the effects of varying the modelling time step on the performance, parameters and structure of hydrological models; (ii) develop a hydrological model operating at different time steps, from daily to sub-hourly, through a unified, robust and coherent modelling framework at different time scales. Our starting point is the chain of conceptual rainfall-runoff models called ‘GR’, developed at Irstea, and in particular the daily ‘GR4J’ lumped model. The GR4J model will be the baseline model to be effectively downscaled up to sub-hourly time steps following a top-down approach. An hourly adaptation of this model had already been proposed in previous research studies, but some questions on the optimality of the structure at sub-daily time steps were still open. This thesis builds on these previous studies on the hourly model and responds to the operational expectations of improving and adapting the model at multiple sub-daily and sub-hourly time steps, which is particularly interesting for flood forecasting applications. For our modelling tests, we built a database of 240 unregulated catchments in metropolitan France, at multiple time steps, from 6-minute to 1 day, using fine time step hydro-climatic datasets available: (i) 6-min rain gauges and higher spatial-density daily reanalysis data for precipitation; (ii) daily temperature data for potential evapotranspiration (making assumptions on sub-daily patterns); (iii) sub-hourly variable time step streamflow data. We investigated the impact of the inputs temporal distribution on model outputs and performance in a flood simulation perspective based on 2400 selected flood events. Then our model evaluation focused on the consistency of model internal fluxes at different time steps, in order to ensure obtaining a satisfactory model performance by a coherent model functioning at multiple time steps. Our model diagnosis led us to identify and test a significant improvement of the model structure at sub-daily time steps based on the complexification of the interception component of the model. Thus, we propose a new version of the model at multiple sub-daily time steps, with the addition of an interception store without extra free parameters. Our tests also confirm the suitability at multiple time steps of a modified groundwater exchange function proposed earlier, leading to overall improved model accuracy and coherence.