In the course of centuries, man has modified the hillslopes in the cultivated parts of the Mediterranean region by creating ditches and terraces that are supported by embankments on the downstream side. As the presence of these man-made landscape features influences storm flow generation, their degradation is a cause of increased flood risk, erosion and eventually crop damage. Distributed hydrological modelling is an important means of estimating the anthropogenic influence in the form of tillage and alteration of the drainage network on storm flow. However, improving distributed storm flow models by parameterisation, calibration and validation is a difficult exercise due to the many spatially distributed parameters involved. This study aims to establish a strategy to parameterise a distributed storm flow model at three different scales: the plot scale at which decisions are made by local farmers, the hillslope scale where multiple plots combine into terraces, and the catchment scale at which water resources are managed by the local water board. The presented methodology combines field experiments with computational modelling. The Roujan experimental catchment in Southern France, which covers 0.91 km2, is divided into no less than 144 plots, has 13 km of terraces and is drained by 9 km of ditches. Discharge, weather and land use data were acquired for storm events since 1992. Two additional stream flow recorders were installed in 2010, bringing the total number of stream flow recorders in the Roujan catchment to eight. As a soft monitoring approach, fourteen stage recorders were installed at various locations, measuring water stage at scales ranging from the plot scale (1500 m2) to the catchment scale (~1 km2). Simulations of storm flow events were performed using MHYDAS, a rainfall-runoff model developed by INRALISAH for studying the effects of agricultural management (Moussa, 2002). A first step in the model parameterisation was to define a strategy based on i) the single-criterion performance, such as the Nash-Sutcliffe coefficient, simulated volume error and simulated peak flow error, and on ii) the multicriteria performance based on the general hydrologic response. The second step was to use these criteria to compare single-event performance, where overland flow celerity, channel runoff celerity, and channel infiltration rate parameters were calibrated per storm event, to multi-event performance, where these parameters were assumed equal for all events. In the last step, stage and stream flow data from multiple locations within the catchment were used to evaluate single-site versus multi-site performance. Results of the comparison of single-event versus multi-event performance using 31 storm events demonstrate that overland flow celerity, channel runoff celerity, and channel infiltration rate have remained relatively constant over a period of 12 years. Comparison of single-site versus multi-site for one storm event shows similar performance between upstream stations, but much less between the station at the outlet and upstream stations. The complementary data supplied by stage recorders allows for significant improvement of the storm flow model.