Résumé
Several hydrological models are currently in use for predicting floods at the catchment outlet. These vary from lumped rainfall-runoff models which ignore flow processes in the catchment to distributed process-based models which attempt to quantify lateral and vertical fluxes. Few of these models integrate extensive field work and most are based almost entirely on readily available digital data. However, flow processes are strongly affected by linear features (paths, roads, hedges...) which are not easily detectable on most digital supports and by factors such as surface roughness or soil depth for which there is often no real data. In the approach described in this thesis, the catchment was subdivided into landscape units based on field mapping. These units represent more or less homogeneous morpho-dynamic units that govern vertical and lateral fluxes of water and soil. The thesis describes the method used for creating the units and landscape segments which will be used for quantifying both runoff and soil erosion.