Exploring the relationship between hydroclimatic stationarity and rainfall-runoff model parameter stability: A case study in West Africa

Forecasting the hydrological impacts of climatic and anthropic changes assumes that the evolution of model parameters under changing conditions can be predicted. Hence it is necessary to study the relationship between hydroclimatic variability and model parameter values. In this paper, we explore this issue by implementing a daily lumped hydrological model (GR4J, Perrin et al. (2003)) on the Upper Oueme watershed (10,050 km(2), Benin). West Africa was subjected to changing climatic and hydrological conditions during the second half of the last century, and changes in the water balance can be evidenced on this watershed. Contrasted periods are extracted from the available 1954-2002 data set, so that hydrological and pluviometric extreme periods can be defined. First, the magnitude of changes in model parameter values under changing conditions are analyzed, using a resampling method (first approach) and within an equifinality context (second approach). It is shown that significant changes in the rainfall- runoff relationship do not induce significant changes in the model's parameter values. A third original approach analyzes the signature of hydroclimatic variability in model performance. Hence a test is defined that uses interannual model efficiency variances to measure performance homogeneity and a resampling test to statistically characterize the calculated results. This test demonstrates the hydrological relevance of the calibrated parameter sets because the more stationary the rainfall- runoff relationship, the more homogeneous the model's performance.