Estimation of rainfall-runoff model parameters based on a large sample of basin characteristics

Estimation of model parameters has been a major problem since the last two decades and if we want to address this problem in the context of ungauged basins it becomes much more challenging. We might think that technological progress in collecting spatialized data (Geographic Information Systems (GIS), Digital Elevation Models (DEM), radars~) offers the possibility to take into account the spatial variability of rainfall and watersheds physical characteristics, in rainfall-runoff models without carrying out any measurement in the field. To use this information, the lumped approach can be easily replaced by the semi-distributed approach. But, does the integration of additional information in a semi-distributed approach effectively improve the performance of flow simulations of this type of models? Moreover, which kind of heterogeneities are to be taken into account to make significant improvements ? In this paper, we deal with the impacts of using physical heterogeneities of basins on the performances of lumped and semi-distributed approaches. Given the requirement of a large database to produce statistically significant results, chimeras basins were used. Chimera basins are built by gathering two real but physically distant basins. Thus, we obtain a large sample of artificial catchments having real rainfall and runoff data at the scale of each real sub-basin.