Is the regionalization of conceptual rainfall-runoff models impossible?

[Departement_IRSTEA]GMA [TR1_IRSTEA]11 - VERSEAU / TRANSFEAU

English. The objective of this paper is to discuss present three reasons why, in our view, the regionalization of conceptual soil moisture accounting (SMA) rainfall-runoff (RR) model parameters is excessively difficult, and perhaps impossible, and to present what may represent a partial solution to the problem of ungauged basins. We first present and discuss the three reasons, which complicate the most, the regionalization of conceptual rainfall-runoff models: 1. overparametrization: the presence of too many unnecessary parameters to calibrate, with too little information contained in precipitation, evaporation and runoff time series, and the unavoidable numerical interdependence of parameters due to the calibration process, contribute to reduce the identifiability of parameters; 2. dependency on input data bias: the parameters of RR models do not represent the behaviour of a watershed in absolute terms as the calibrated value of RR model parameters is conditional to the systematic biases present in the input data series: evaporation and precipitation; 3. the paradoxal lack of systematic link between parameter precision and model efficiency. We consider that the three above reasons are enough to worsen quite considerably the task of those who seek a relevant regionalization of RR model parameters. Based on their understanding, we see three possible directions for further work: first, we consider that the most needed action is to turn the present RR models more parsimonious; second, we should try to assess the necessary amount of information needed to get a sufficiently good estimate of parameters. Partial results from an ongoing Ph.D. thesis are presented; third, we believe that there are some possibilities to transfer information between models and use the synergy of different modelling approaches.