Improving the temporal transposability of lumped hydrological models on twenty diversified U.S. watersheds

•Individual and collective performances of twenty lumped conceptual models.•Calibration/validation process on contrasted conditions on twenty watersheds.•Deterministic multimodel approaches increase performances.•Weight averaged multimodel enhances transposability.•No fundamental link between individual performance and multimodel weights.

Study regionTwenty diversified U.S. watersheds.Study focusIdentifying optimal parameter sets for hydrological modeling on a specific catchment remains an important challenge for numerous applied and research projects. This is particularly the case when working under contrasted climate conditions that question the temporal transposability of the parameters. Methodologies exist, mainly based on Differential Split Sample Tests, to examine this concern. This work assesses the improved temporal transposability of a multimodel implementation, based on twenty dissimilar lumped conceptual structures and on twenty U.S. watersheds, over the performance of the individual models.New hydrological insights for the regionIndividual and collective temporal transposabilities are analyzed and compared on the twenty studied watersheds. Results show that individual models performances on contrasted climate conditions are very dissimilar depending on test period and watershed, without the possibility to identify a best solution in all circumstances. They also confirm that performance and robustness are clearly enhanced using an ensemble of rainfall-runoff models instead of individual ones. The use of (calibrated) weight averaged multimodels further improves temporal transposability over simple averaged ensemble, in most instances, confirming added-value of this approach but also the need to evaluate how individual models compensate each other errors.