Comparison of calibration strategies of a conceptual hydrological model and their influence on streamflow simulation performance, snow cover representation and parameter identifiability.

Having a realistic estimate of snow cover by conceptual hydrological models is a great challenge faced by hydrologists for flood prediction using these models. Moreover, calibrated model parameters are another source of uncertainties in climate change impact studies. The objective of this study is to improve the calibration of the conceptual hydrological model GR4J with the snowmelt module Cemaneige, in order to obtain a more realistic simulation of the snow cover and to reduce the uncertainty of the free parameters. The performance of the two models was tested on 12 watersheds with a natural hydrological regime in southern Quebec, Canada. In the first two strategies, all parameters were calibrated against observed streamflow only, using a local and a global algorithm. Then, snow water equivalent (SWE) observations from a permanent monitoring network were used to independently calibrate the snow parameters in the Cemaneige model. Finally, multi-objective optimization was used to simultaneously calibrate hydrological and snow parameters. Results show that the inclusion of snow survey data within a multi-objective optimization algorithm, where the parameters of both snow and hydrological models are calibrated against SWE and discharge observations, improved the simulation of the SWE as well as the identifiability of the parameters. A set of equifinal parameters recorded by the algorithms during optimization and which give the same optimal performance were used to quantify the uncertainty generated by equifinality on the detection of climate change impact on streamflow.