Résumé
Climate change impact on the groundwater resource is usually assessed based on a modeling approach. Bias-corrected outputs of regional climate models (RCM) are used to force an hydrological model on a control and a scenario period. The climate change impact on the hydrological variables is investigated by comparison of the output of the hydrological model for the control and scenario runs (Bloeschl and Montanari, 2010). The reliability of the resulting climate change impact prediction is dependent on the uncertainties that stem from all steps of the modelling sequence. In particular, serious issues are associated with downscaling to the local, hydrological scale (Teutschbein et al., 2011). This work investigates the effect of the non-linearity of the hydrological model on the prediction of the climate change impact on the hydrological model output, for a small-scale mediterranean karst system. The analysis is performed for the Lez spring system (southern France). The Lez spring is one of the largest french karst spring. It is used as the major water resource for the city of Montpellier (250 000 inhabitants). Prediction of the spring discharge (during high water periods) and of the water level within the spring conduit (during low water periods) under different resource use and climate scenarios is of high importance for the sustainable management of the resource. First, we test the ability of control run simulations to reproduce the spring discharge and the spring water level distributions that result from rain-gauge based simulations, for the 1980-2010 control period. Then, a linear perturbation is applied to the control and the observed rainfall time series. We compare the change in the hydrological model output distribution for the perturbed control and observed rainfall time series.