A comparison of 1701 snow models using observations from an alpine site

There are many models that attempt to predict physical processes in snow on the ground for a range of applications, and evaluations of these models show that they have a wide range of behaviours. A review of snow models, however, shows that many of them draw on a relatively small number of process parameterizations combined in different configurations and using different parameter values. A single model that combines existing parameterizations of differing complexity in many different configurations to generate large ensembles of simulations is presented here. The model is driven and evaluated with data from four winters at an alpine site in France. Consideration of errors in simulations of snow mass, snow depth, albedo and surface temperature show that there is no “best” model, but there is a group of model configurations that give consistently good results, another group that give consistently poor results, and many configurations that give good results in some cases and poor results in others. There is no clear link between model complexity and performance, but the most consistent results come from configurations that have prognostic representations of snow density and albedo and that take some account of storage and refreezing of liquid water within the snow.