Performance and reliability of multimodel hydrological ensemble simulations based on seventeen lumped models and a thousand catchments

摘要

This work investigates the added value of ensembles constructed fromseventeen lumped hydrological models against their simple averagecounterparts. It is thus hypothesized that there is more informationprovided by all the outputs of these models than by their single aggregatedpredictors. For all available 1061 catchments, results showed that the meancontinuous ranked probability score of the ensemble simulations were betterthan the mean average error of the aggregated simulations, confirming theadded value of retaining all the components of the model outputs.Reliability of the simulation ensembles is also achieved for about 30% ofthe catchments, as assessed by rank histograms and reliability plots.Nonetheless this imperfection, the ensemble simulations were shown to havebetter skills than the deterministic simulations at discriminating betweenevents and non-events, as confirmed by relative operating characteristicscores especially for larger streamflows. From 7 to 10 models are deemedsufficient to construct ensembles with improved performance, based on agenetic algorithm search optimizing the continuous ranked probability score.In fact, many model subsets were found improving the performance of thereference ensemble. This is thus not essential to implement as much asseventeen lumped hydrological models. The gain in performance of theoptimized subsets is accompanied by some improvement of the ensemblereliability in most cases. Nonetheless, a calibration of the predictivedistribution is still needed for many catchments.