ABSTRACT In areas modified by human activities, accelerated erosion can lead to serious environmental issues such as the flooding of urban areas or the pollution of water bodies. In order to prevent or to mitigate such events, it is therefore necessary to be able to predict the dynamic as well as the spatial extent of runoff production and particle transfer. To meet this demand and to reflect the complexity of the processes involved as well as the spatial heterogeneity of the landscape, several modelling approaches of various complexities have been developed. Many research efforts are devoted to the development of physically-based models able to improve our understanding and modelling of these fluxes. One of the main obstacles to the application of such models is the difficulty to describe the spatial and temporal variability of the input parameters. In this context, the objective of this study is to develop a robust physically based modelling approach and to adapt its parameterisation to be able to incorporate coarse scale input parameter. A dynamic erosion model coupling the Shallow Water equations with the Hairsine-Rose model was developed. It uses a well-balanced numerical scheme with a hydrostatic reconstruction to preserve the equilibrium and the positivity of water height and to be able to capture spatial heterogeneities. A first application is realised at the plot scale to test the model predictive ability. In a second step, to integrate inner grid variability when modeling at the catchment scale, for each cell, we use the proportion of wetted area as a microtopography indicator. For the case of erosion, the system is coupled to the sediment transport equations. In such context, an additional equation describing the micro-topography evolution caused by erosion is introduced and the numerical scheme of Godunov-type for this model is implemented. The results of a second application at the catchment scale are presented.