Flash floods are a feature of the Mediterranean catchments. They are caused by extreme rainfall events, especially in autumn. Improving our outstanding about their generation and propagation effects in large-scale catchments is therefore of great importance. In order to progress in flood modelling and understanding of large catchments, it is interesting to couple hydrological and hydraulic models in order to consider river propagation effects on the simulated hydrographs. The objective of this study is two-fold: to create a new distributed hydrological model, built from a data-based approach, within the JAMS modelling framework and to proceed with data coupling with the MAGE 1D hydraulic model. In contrast to many hydrological models that have been developed over the last 30 years based on bottom-up or reductionist approach, we use in this research a combination of data analysis and model conceptualization (top-down approach). Based on this, we developed a simple distributed model in JAMS modeling platform following Kirchner’s methodology (Kirchner, 2009) that has three parameters derived from available data. The distributed hydrological model is based on the hypothesis that catchment behaviour can be represented as a non-linear reservoir model, assuming that discharge at the outlet is only a function of catchment storage. The approach was successfully tested in the Ardèche catchment (2355 km2), located in the southern part of France, characterized by a high degree of heterogeneity and variability of geology, pedology and slopes. The Mediterranean type climate is also contrasted, with seasonal heavy rainfall events during autumn and dry summers. The parameters of the simple model are estimated at the gauged locations and regionalization is done according to geology. The catchment is discretized into sub-catchments of about 10 km2. The model is forced using hourly rainfall and evapotranspiration data as primary inputs. We coupled our distributed hydrological model and hydraulic MAGE 1D model by external communication (data coupling) where outputs from hydrological model become inputs to hydraulic model without defining any internal boundary condition between them. Outputs are discharge rates in the reach network that are transferred into the MAGE model as either lateral flows (coming from adjacent land) and/or local inflows. We run the hydrological simulations for the whole examined period (2000-2012). On another side, the coupled model with MAGE was run continuously at the hourly time step for chosen extreme events (year 2003, 2008 and 2011). The results highlight that it is possible to connect those two models and that the timing and magnitude of simulated discharge is better reproduced by the data coupled model than by hydrological model. We show that for better peaks representation in downstream catchments during the flood events, coupling of the models appear to be a reliably robust approach due to the more efficient routing scheme used in the coupled model as compared to the use of the standard routine module of JAMS.