A review of efficiency criteria suitable for evaluating low-flow simulations

► We review and compare several criteria used for the evaluation of low-flow simulations. ► We propose a new criterion for the evaluation of low-flow simulations. ► We evaluate the relevance of this criterion on a large set of catchments showing various low flow characteristics. Low flows are seasonal phenomena and an integral component of the flow regime of any river. Because of increased competition between water uses, the demand for forecasts of low-flow periods is rising. But how low-flow predictions should be evaluated? This article focuses on the criteria able to evaluate the efficiency of hydrological models in simulating low flows. Indeed, a variety of criteria have been proposed, but their suitability for the evaluation of low-flow simulations has not been systematically assessed. Here a range of efficiency criteria advised for low flows is analysed. The analysis mainly concentrates on criteria computed on continuous simulations that include all model errors. The criteria were evaluated using two rainfall–runoff models and a set of 940 catchments located throughout France. In order to evaluate the capacity of each criterion to discriminate low-flow errors specifically, we looked for the part of the hydrograph that carries most of the weight in the criterion computation. Contrary to what was expected, our analysis revealed that, in most of the existing criteria advised for low flows, high flows still make a significant contribution to the criterion’s value. We therefore recommend using the Nash–Sutcliffe efficiency criterion calculated on inverse flow values, a valuable alternative to the classically used criteria, in that on average it allows focusing on the lowest 20% of flows over the study period.