Challenging Large-scale Hydrological Simulations with Streamflow Observations: Response versus Persistence

Abstract:

Land surface models and large-scale hydrological models are often used to study climate change impacts on hydrology at regional to global scales. These impacts are then presented as maps of change in specific runoff metrics that are relevant to basin management and water resources planning. Knowing the limits of model performance for the respective metrics of interest at different spatial and temporal scales is thus important, but often performance is only known for annual or long-term means. This contribution summarizes and reflects on the challenge of continental hydrological model simulations from the WATCH multi-model ensemble with distributed streamflow observations from small basins of reference networks in Europe. Characteristics of hydrological dynamics that were compared include spatial and temporal runoff persistence, high and low flows, and long-term trends and variability. Whereas common annual statistics between models and observations correlate well even if the amounts disagree, larger differences were found for metrics that focus on the dynamics of streamflow response and persistence. For example, models appear to respond comparably fast to precipitation, and as a consequence underestimate the duration of streamflow drought events. Investigating the general streamflow persistence in time and space, however, also showed large differences among the different models. Long-term trends in annual flow and annual weekly peak flow in Europe agreed on the large-scale patterns, but particularly seasonal trends and trends in extremes in regions with mixed observed runoff trends or in complex terrain revealed discrepancies to the observations even regarding the sign of the trend. Before the display of changes in hydrological characteristics related to response and persistence of flow situations, models should therefore always be tested specifically for their limits to represent such metrics.