Comparing Large-Scale Hydrological Model Simulations to Observed Runoff Percentiles in Europe

Abstract Large-scale hydrological models describing the terrestrial water balance at continental and global scales are increasingly being used in earth system modeling and climate impact assessments. However, because of incomplete process understanding and limits of the forcing data, model simulations remain uncertain. To quantify this uncertainty a multimodel ensemble of nine large-scale hydrological models was compared to observed runoff from 426 small catchments in Europe. The ensemble was built within the framework of the European Union Water and Global Change (WATCH) project. The models were driven with the same atmospheric forcing data. Models were evaluated with respect to their ability to capture the interannual variability of spatially aggregated annual time series of five runoff percentiles—derived from daily time series—including annual low and high flows. Overall, the models capture the interannual variability of low, mean, and high flows well. However, errors in the mean and standard deviation, as well as differences in performance between the models, became increasingly pronounced for low runoff percentiles, reflecting the uncertainty associated with the representation of hydrological processes, such as the depletion of soil moisture stores. The large spread in model performance implies that any single model should be applied with caution as there is a great risk of biased conclusions. However, this large spread is contrasted by the good overall performance of the ensemble mean. It is concluded that the ensemble mean is a pragmatic and reliable estimator of spatially aggregated time series of annual low, mean, and high flows across Europe.