Transferability and parameter uncertainty of hydrological models for estimating future mean and extreme discharges in the context of climate change

Previous studies on climate change impacts have paid considerable attention on assessing the uncertainties associated with greenhouse gas emission scenarios and General Circulation Model (GCM) structures. Increasing studies stress the need for routinely testing the performance and analyzing uncertainty of hydrological models in the impact assessment. The overarching objectives of this study are 1) to investigate the transferability of the hydrological model parameters to climatic conditions that are different from that in the calibration period, and 2) to compare the uncertainties in the future mean and extreme river discharges due to the equifinality of model parameters and the choice of calibration periods. A lumped Xinanjiang Hydrological Model of the Huai River Basin in China is used to test the methodology. The transferability of model parameters is tested in the context of historical climate variability using the differential splitsample test. Four GCMs participating in the CMIP5 data portal are selected. The results show that the transferability of the parameters calibrated from a wet period to a dry period is poorer than the other way around. The model error as well as the variability in the simulation due to equifinality increase with the increase in the difference in rainfall amounts between the calibration and validation periods. Generally, the uncertainty due to the choice of calibration periods takes larger share of the total parameter uncertainty in the projected future mean discharge. When the calibration period contains enough information of climate variability, the equifinality becomes the main source of parameter uncertainty for high-return-period extreme discharge. The results will provide the basis for better understanding the uncertainties in assessing hydrological impacts of climate change.