Geophysical Research Letters | Mudryk et al. 
The relationship between land surface temperature and snow cover extent trends is examined in three distinct types of ensembles over the 1981-2010 period: an observation-based ensemble, a representative selection of CMIP5 coupled climate model output, and two large initial condition coupled climate model ensembles. Observation-based estimates of snow cover sensitivity are stronger than simulated over midlatitude and alpine regions. Observed sensitivity estimates over Arctic regions are consistent with simulated values. Anomalous snow cover extend trends present in one dataset, the NOAA climate record, obscure the relationship to surface temperature seen in the rest of the analyzed data. The spread in modeled snow cover trends reflects roughly equal contributions from inter-model variability and from natural variability. Together, the anomalous relationship between surface temperature and snow cover expressed in the NOAA climate record and the large influence of natural variability present in the simulations highlight the importance of ensemble-based approaches.
Full text can be found here.