Large Sample Basin Experiments for Hydrological Model Parameterization: Results of the Model Parameter Experiment–MOPEX. IAHS Publ. 307, 2006. 246 How informative is land-cover for the regionalization of the GR4J rainfall–runoff model? Lessons of a downward approach LUDOVIC OUDIN 1 , VAZKEN ANDRÉASSIAN 2 , CÉCILE LOUMAGNE 2 & CLAUDE MICHEL 2 1 Université Pierre-et-Marie Curie, UMR 7619 Sisyphe, Case 105, 4 Place Jussieu, 75252 Paris cedex 5, France ludovic.oudin@ccr.jussieu.fr 2 Cemagref, Hydrology and Water Quality Research Unit, PB 44, 92163 Antony cedex, France Abstract Prediction on ungauged basins is a big issue for operational hydrol- ogy and a challenge for scientists. For regionalization objectives, the first step of the development of a priori parameter estimation is to look for correlations between calibrated parameters and physical catchment descriptors. This article investigates the relationships between the parameters of the GR4J rainfall– runoff model and catchment vegetation characteristics over a large sample of 221 French catchments. First, the possible links between the calibrated parameters of the GR4J model and catchment vegetation types are investi- gated by linear regression. Then, we try to improve these relationships by introducing a more detailed description of the evapotranspiration process, explicitly taking into account vegetation types, following a downward approach. Results show that the GR4J model parameters cannot be determined directly from vegetation characteristics, and that the situation is not improved by a more detailed approach to evapotranspiration modelling. Key words downward approach; evapotranspiration; land use; rainfall–runoff model; regionalization INTRODUCTION Vegetation type is one of the most often cited driving variables of catchment behaviour. This consensus has its roots in the numerous studies implemented by forest hydrologists on small catchments during the 20th century. Several reviews are avail- able on this topic: see for example, Bosch & Hewlett (1982) or Andréassian (2004). Mainly based on paired catchment experiments, these studies consisted of deforesta- tion and reforestation experiments, by which it was possible to demonstrate without doubt that forest cover could have an important role in the water balance at the catchment scale. However, the fact that vegetation has a role in a water cycle does not necessarily imply that vegetation is informative for regionalization objectives. Indeed, to use land cover for regionalization applications, we must be able: (i) to quantify its impact on the water cycle at the catchment scale; and (ii) to isolate its impact from other linked factors: soils, climate. At first sight, the most rational approach to explicitly introduce vegetation characteristics into hydrological models is to use a mechanistic approach, with a physically-based model whose parameters are directly linked with vegetation types. Several large-scale experiments have supported the development of Soil-Vegetation-