Integrated implementation of the Raven HBV-EC model and describing uncertainty in a multimodel ensemble evaluation

The average weekly streamflow simulations over the period 2000–2020 for the Au Saumon station are shown in Figure 5, where the pale and dark blue coverages illustrate the distribution of the streamflow ensemble for 48 the models (5 to 95% and 25 to 75%, respectively), while the red line illustrates the median flow and the black line, the observed flow. [...] This evaluation involved comparing the RMSE of the best model for each of the different hydrometric stations (this model thus differs from on site to the other, out of the pool of 48) to the RMSE values of the four multimodel sets. [...] It serves as a standardized assessment of the root mean square error, computed as the difference between the RMSE of the best model and the RMSE of the different sets, divided by the RMSE of the best model. [...] Following the selection of a subset of models, and exploring the model parameter uncertainty, the subsequent phase of the study involved introducing noise to the input data to assess uncertainty within the multimodel ensembles. [...] The results in terms of the reliability diagram in Famine watershed illustrated that using a pool of the combination of the 12 models for different percentages of the noises, improved the reliability comparing with using the initial Raven HBV-EC model.