The present thesis aims to work out the general characteristics of the sensitivity of numerical models of flow within karst systems. A special attention is devoted to the study of the influence of karst specificities (high heterogeneity, duality of flow, highly non-linear behaviour) on the sensitivity propagation, with the final purpose of answering the following questions: (i) is it possible to calibrate the model ? (ii) is the calibration robust ? (iii) is it possible to reduce equifinality, through multi-objective calibration or through multi-variable calibration ? The analysis is performed for global reservoir models and distributed, hybrid flow models. This contribution stresses the potentialities of local sensitivity analyses. Despite their inherent limitations (local approximation), local analyses have proved to bring valuable insights into the general behaviour of complex, non-linear flow models, at little computational cost. Besides, this contribution also stresses the interest of multi-variable calibration as compared to multi-objective calibration, as regards equifinality reduction.
