A Bayesian multi-model framework for structure selection and parameter estimation for lumped parameter modeling in karst hydrology.

Lumped parameter modeling in karst hydrology has been widely developped in the last decades. Uncertainty in model conceptualization, which often leads to one unique model structure, is frequently neglected. This issue is particularly important for karst hydrology, where hydrological systems are highly heterogenous and information about the structure is difficult to obtain. In this work, we delopped a Bayseian multi-model framework that allows to calibrate simulteanously a set of model structure and associated parameters. This constitutes a significant step forward compared with classical calibration approaches that allow (i) to provides an ensemble of predictions considering both structural and parametric uncertainties, and (ii) to avoid epistemic error due to model structure selection, wich is generally influenced by the subjective conceptualization of the karst hydrological system by the modeler. The methodology is illustred with a Bayesian inference procedure among a large range of lumped parameter model structure considered for the simulation of discharge at fontaine de Vaucluse karst spring (southern France).