Benchmarking hydrological models for low-flow simulation and forecasting on French catchments

摘要

Low-flow simulation and forecasting remains a difficult issue forhydrological modellers, and intercomparisons can be extremely instructive forassessing existing low-flow prediction models and for developing more efficientoperational tools. This research presents the results of a collaborativeexperiment conducted to compare low-flow simulation and forecasting modelson 21 unregulated catchments in France. Five hydrological models (fourlumped storage-type models – Gardenia, GR6J, Mordor and Presages – and onedistributed physically oriented model – SIM) were applied within a commonevaluation framework and assessed using a common set of criteria. Two simplebenchmarks describing the average streamflow variability were used to setminimum levels of acceptability for model performance in simulation andforecasting modes. Results showed that, in simulation as well as inforecasting modes, all hydrological models performed almost systematicallybetter than the benchmarks. Although no single model outperformed all theothers for all catchments and criteria, a few models appeared to be moresatisfactory than the others on average. In simulation mode, all attempts torelate model efficiency to catchment or streamflow characteristics remainedinconclusive. In forecasting mode, we defined maximum useful forecastinglead times beyond which the model does not bring useful information comparedto the benchmark. This maximum useful lead time logically varies betweencatchments, but also depends on the model used. Simple multi-modelapproaches that combine the outputs of the five hydrological models weretested to improve simulation and forecasting efficiency. We found that themulti-model approach was more robust and could provide better performancethan individual models on average.