Revealing the positive influence of young water fractions derived from stable isotopes on the robustness of karst water resources predictions

Introducing additional information sources, such as hydrochemical signatures and water isotopes, into the model calibration has shown to be useful to enhance model robustness by increasing parameter identifiability and maintaining simulation reliability. Our study explores the added value of discharge young water fractions (Fyw, derived from the volume-weighted delta 18O concentrations) on the model reliability as a calibration constraint. For this, we coupled a karst hydrological model (VarKarst) with a catchment-scale transport model (StorAge Selection (SAS) function approach) to simulate discharge delta 18O concentration (delta 18OQ) and corresponding Fyw. We performed a multi-variable calibration scheme by simultaneously constraining a large model ensemble (1x106 realizations) with respect to the model performance on discharge and Fyw. By searching a model output space in which the model performance on discharge and model performance on Fyw provides an optimal trade-off, we extracted hydrologically more informative model realizations. We tested our calibration approach at the Wasseralm spring, which supplies drinking water to the city of Vienna, Austria. The contribution of the information content of Fyw to the model robustness was assessed by the degree of reduction in the parameter and simulation uncertainties. Our results indicate that the inclusion of Fyw notably reduced the uncertainty in model parameters (6 parameters out of 8), simulations (14 % vs. 10 % by delta 18OQ), and water balance components for the model internal states (around 40 %). Our findings reveal that Fyw confirms the model reliability (KGE: 0.71 +/- 0.01 in validation) in that it mainly reduces the parameter equifinality by providing physically more plausible and identifiable parameter sets. Therefore, Fyw is a potentially useful metric to better constrain the model outputs, thereby limiting the model uncertainty.