Abstract. This paper investigates issues involved in calibrating hydrological models against observed data when the aim of the modelling is to predict future runoff under different climatic conditions. To achieve this objective, we tested two hydrological models, DWBM and SIMHYD, using data from 30 unimpaired catchments in Australia which had at least 60 yr of daily precipitation, potential evapotranspiration (PET), and streamflow data. Nash-Sutcliffe efficiency (NSE), modified index of agreement (d 1 ) and water balance error (WBE) were used as performance criteria. We used a differential split-sample test to split up the data into 120 subperiods and 4 different climatic sub-periods in order to assess how well the calibrated model could be transferred different periods. For each catchment, the models were calibrated for one sub-period and validated on the other three. Monte Carlo simulation was used to explore parameter stability compared to historic climatic variability. The chi-square test was used to measure the relationship between the distribution of the parameters and hydroclimatic variability. The results showed that the performance of the two hydrological models differed and depended on the model calibration. We found that if a hydrological model is set up to simulate runoff for a wet climate scenario then it should be calibrated on a wet segment of the historic record, and similarly a dry segment should be used for a dry climate scenario. The Monte Carlo simulation provides an effective and pragmatic approach to explore uncertainty and equifinality in hydrological model parameters. Some parameters of the hydrological models are shown to be significantly more sensitive to the choice of calibration periods. Our findings support the idea that when using conceptual hydrological models to assess future climate change impacts, a differential split-sample test and Monte Carlo simulation should be used to quantify uncertainties due to parameter instability and non-uniqueness.

No abstract