Study of parameter stability of a lumped hydrologic model in a context of climatic variability

Central and West Africa were affected by an often marked reduction in rainfall and runoff around the year 1970. Has the behaviour of the catchments in these regions been changed as a result? Seventeen basins are used in this study, and are characterised by stationary or non-stationary annual rainfall or runoff time-series. An approach based on lumped hydrological modelling with a monthly time step (GR2M water balance model) and automatic parameter calibration is used to try to answer the question. Parameter stability of the models calibrated before and after the occurrence of possible rainfall or runoff deficit is analysed using estimations of confidence region. Minimisation of the least squares objective function provides a local optimum around which confidence regions are estimated in a non-linear context. The volumes of indifference represented by the confidence regions are analysed by their cross-sections on the planes defined by the three parameters of the model taken in pairs. For each basin, the cross-sections relative to different periods of calibration are interpreted in terms of possible parameter stability. This study shows that there is no link between parameter stability and the stationary behaviour of rainfall or runoff series of some catchments. (C) 2003 Elsevier Science B.V. All rights reserved.