Robustness of hydrological models for simulating impacts of climate change on high and low streamflow

• Robustness of HBV and GR6J evaluated for simulating high and low streamflow. • Evaluation on historic periods that resemble projected future climatic conditions. • Models performed well partly due to the use of a multi-objective function for calibration. • Future streamflow uncertainty stems mainly from uncertainties in climate scenarios. Hydrological models are used to simulate the rainfall-runoff transformation to quantify climate change impacts on extreme streamflow. However, parameters of hydrological models optimized for historic conditions may not be valid for future scenarios. Few studies developed advanced approaches for testing validity of models for climate change impact assessments. This study developed a framework for evaluating the robustness of models for simulating climate change impacts on high and low streamflow. The innovation of this framework lies in using future climatic timeseries directly to select historic years for calibration and validation. The framework was applied to two hydrological models (HBV and GR6J) in the Lesse catchment, Belgium. Models were evaluated on historic periods that resemble climatic conditions projected by future climate scenarios. Both models showed a loss in performance in validation periods that resemble future conditions compared to calibration periods with historic conditions. However, both models still performed acceptable, possibly due to the use of a multi-objective function for calibration. Climate change causes a median change in annual maximum daily discharge between −14 % and +27 % in 2100 compared to 1991–2020. The median projected change in annual minimum 7-day mean discharge was between −66 % and +13 %. The uncertainty stemmed mainly from future greenhouse gas emissions and climate response, but uncertainty subject to model structures and calibration approaches cannot be neglected. The ability of hydrological models to simulate climate change impacts should not be taken for granted. Improving model structures may enhance robustness and therefore contribute to climate change impact projections in the future with less uncertainty.