How robust is a multi-model ensemble of conceptual hydrological models to climate change?: lessons from 44 models in 582 river basins.

Conceptual hydrological models are used to understand and predict the runoff process in river basins, which contributed to decision-making in disaster management. It is important to develop conceptual hydrological models robust to changing climate to apply them to the climate change risk assessment. However, many previous works pointed out that conceptual hydrological models which are calibrated using the current climate data may provide inaccurate prediction of runoff processes in the future climate. It is a grand challenge in hydrology to realize highly robust runoff prediction to climate change using conceptual hydrological models. Although multi-model ensemble (MME) has been recognized as an promising method to improve the robustness of prediction based on conceptual hydrological models, the robustness of MME of conceptual hydrological models has yet to be thoroughly evaluated. In this study, we evaluated the robustness of MME by using 44 conceptual hydrological models in 582 river basins of the United States. We performed parameter calibration for each model in each river basin and validated the performance of each model as well as MME in the period which is different from the calibration period. We found that the MME prediction was more accurate and robust than the prediction of each model alone during the validation period. This superiority of MME becomes more apparent in the river basins where the climatology of river discharge in the validation period is more different from that in the calibration period, which implies the robustness of MME to climate change. However, it is difficult to quantify this robustness of MME when the number of basins and models is small (e.g., 10 basins and 5 models). The result suggests that the large number of samples is crucially needed to evaluate the robustness and uncertainty in hydrological prediction.