Résumé
<p>Modelling snow and glacier melt runoff processes is essential for local water supply, hydropower management and flood forecasting. However, melt runoff modelling in mountainous regions faces two challenges: scarcity of meteorological data and uncertainty in parameter calibration due to limited understanding of the complex hydrological processes. Numerous free parameters have been introduced in most of the snow-accounting routines (SAR) used in operational hydrology in order to compensate for errors in the input data, adapt to local snow processes or deal with scale effects between different snow data.  This talk investigates the degree of liberty required in a SAR to simulate local snow dynamics, distributed snow cover areas and flows at the catchment outlet. To address this issue, a SAR on top of the GR4J model was tested on a dataset covering 17 mountainous catchments (45 to 3580 km²) in the French Alps and Pyrenees. Model calibration and control were based on streamflow series, fractional snow-covered areas (FSC) computed from MODIS snow products and at least one chronicle of local measures of snow water equivalent (SWE) acquired in each catchment over the period 2004−2016. The SAR was applied according to elevation bands of 100 meters and various parametrisation: from ten free parameters (precipitation orographic correction, temperature lapse rate, seasonal variation of the temperature lapse rate, snowfall gauge under-catch correction, thermal inertia of the snow pack, melt degree day factor, variable melt factor, ice degree day factor, 2-parameter hysteresis between SWE and FSC) to only one free parameter (snowfall gauge under-catch correction). Results shows that the one-free-parameter SAR is as efficient as more free structures to simulate both distributed and local snow dynamics as well as runoff. Interestingly, using a SAR without any free parameters by fixing the snowfall gauge under-catch correction to a value of 150% leads only to a deterioration of local SWE dynamics. These findings suggest that it is possible, and even advisable, to limit the number of free parameters in temperature-index models in order to reduce problems of over-parameterisation and equifinality.</p>