Forecasting the formation of critical snow layers using a coupled snow cover and weather model

The snow cover model SNOWPACK simulates snow cover formation and evolution based on meteorological data. In the past, these data were measured by automated weather stations. Recently, SNOWPACK was also forced with data from numerical weather prediction models (NWP). In this study we assess the capability of such a model chain to simulate two types of critical layers, i.e. surface hoar and melt-freeze crusts, for virtual north and south-facing slopes as well as a flat study plot. Meteorological key parameters for snow cover formation and evolution, e.g. precipitation, radiation, air temperature and relative humidity, were measured and compared to the forecasted data to evaluate the performance of the NWP model. Systematic errors of the NWP model were corrected and the adjusted data were used to force SNOWPACK. The formation of 80 critical layers - 35 surface hoar layers and 45 melt-freeze crusts - from seven winters between 2005 and 2012 observed in the Columbia Mountains of British Columbia, Canada were investigated. To assess the performance of SNOWPACK at simulating the presence and absence of critical layers on different aspects, monthly manual snow profiles systematically observed on north, south and flat terrains between January and March during the winter of 2010-2011 were compared to the corresponding snow cover simulations. The overall accuracy to predict the formation of melt-freeze crusts was found to be between 65% and 77% depending on the aspect. Surface hoar layers were modeled with accuracy between 89% and 100%. The presence and absence of critical layers within the snow cover - surface hoar and melt-freeze crusts - were investigated during the winter of 2010-2011. Ten out of eleven observed critical layers within the snow cover were simulated, resulting in an accuracy of 91%. However, some surface hoar layers and melt-freeze crusts were simulated but not observed resulting in a low accuracy for the absence of critical layers. The simulated snow height tended to be under-estimated during the early winter season and over-estimated during the late winter season for both slopes and the flat site, especially in March on the south-facing aspect. Nevertheless, the model chain is promising considering the source of the input data. This study showed that such a model chain could become a useful tool for avalanche warning services in the future, especially for data sparse areas. (C) 2013 The Authors. Published by Elsevier B.V. All rights reserved.