Rainfall-runoff modelling of a rocky catchment with limited data availability: Defining prediction limits

A distributed model was used to simulate the main hydrological processes in the Ourthe catchment, situated in the southern hilly part of Belgium. The study revealed important uncertainties associated with the current modelling, particularly with regard to the data of the groundwater system. These are likely to influence the global performance of the numerical model and as such a joint deterministic-stochastic protocol based on Monte Carlo simulations was used to calibrate the model and to assess its current prediction limits. Despite the lack of appropriate time series of observations of the groundwater system, the model performance was assessed not only on the overall simulation of streamflow but also on the prediction of hydrograph subflows for which a digital recursive filter was developed and used. The subflow estimates revealed the importance of the groundwater system as a main contributor to the streamflow. Correspondingly, the deterministic-stochastic analysis showed that the sensitivity of predictions to the horizontal saturated hydraulic conductivity of the modelled groundwater system is highest. In general, broad prediction limits were obtained. The model had some difficulties to mimic well the low flows as well as flows above 60m³s^-1. These are believed to be the consequence of the significant data uncertainties and errors in boundary conditions, among other aspects. Nevertheless, the study revealed that considering the subflow estimates for evaluating model performance constitutes a tougher evaluation condition offering the potential of enhancing specificity in the assessment of the model prediction limits.