Assessment of the uncertainty of snowpack simulations based on variance decomposition
Abstract. State of the art numerical snow models essentially rely on observational data for initialization, forcing, parametrization and validation. Such data are available in increasing amount, but the inherent propagation of related uncertainties on the simulation results has received rather limited attention so far. Depending on their complexity, even small errors can have a profound effect on simulations, which dilutes our confidence in the results. This paper quantifies the fractional contributions of some archetypical measurement uncertainties on key simulation results in a high Arctic environment. The contribution of individual factors on the model variance, either alone or by interaction, is decomposed using Global Sensitivity Analysis. The work focuses on the temporal evolution of the fractional contribution of different sources on the model uncertainty, which provides a more detailed understanding of the model's sensitivity pattern. The decompositions demonstrate, that the impact of measurement errors on calculated snow depth and the surface energy balance components varies significantly throughout the year. Some factors show episodically strong impacts, although there overall mean contribution is low while others constantly affect the results. However, these results are not yet to be generalized imposing the need to further investigate the issue for e.g. other glaciological and meteorological settings.