On the need for benchmarks in hydrological modelling

Hydrological models are applied frequently to scientific or practical problems. For many applications it is concluded that the model has been able to reproduce the measurements with ‘acceptable accuracy’. The question is what we mean by this term; the meaning of ‘acceptable accuracy’ can be quite subjective. We might compute statistical goodness-of-fit measures such as model efficiency, but even the use of such a measure does not necessarily allow an objective judgment of model performance. Does an efficiency of 0.8 for the runoff simulations indicate good or poor model performance? The answer depends on whom you ask. But it also depends on what could be achieved given the specific catchment and the observed data. What might be a poor fit for a watershed with excellent measurements might rightly be considered to be good in a watershed where the available data are of poor quality. To truly assess model performance, it is important to compare one’s results with results obtained in some other way, i.e. to choose an appropriate benchmark series.