Résumé
Model validation is still a crucial issue in hydrology. It is usually limited to comparing simulated and measured streamflows, while many other fluxes and storages are also simulated. This is especially true for spatially distributed models, for which multivariable and multiscale validation procedures are needed. Such an approach has been tested on the Fecht research basin (Vosges Massif, France) over a 5-year period, using the semidistributed conceptual MC model, whose physical soundness has been improved. The catchment discretization into land units was based on simple physiographic criteria, which were statistically validated. Water balance parameters were derived from measured soil water retention, without any calibration. Catchment streamflow was used to calibrate water transfer parameters over a single year and validate the model. Two subcatchments were used for the internal validation, and two neighboring nested catchments were used for the regional transposition of both the model and its distributed parameterization. At the mesh scale (250 to 1000 m), snowpack and soil water storages and contribution to streamflow were validated by using measurements from two small catchments. Despite the very limited calibration, the results were generally satisfactory. Nash-Sutclifife efficiencies on daily streamflow varied from 0.92 in calibration to 0.89 in validation and 0.71 in transposition. The results of the simulated patterns of fluxes and storages can therefore be considered with some confidence.