A new continuum approach to snowdrift modelling is introduced. In addition, numerical studies are carried out to identify the influence of time-varying wind conditions on snowdrift simulations. We compare the snowdrift patterns at Grimming mountain, Austria, derived using a time-averaged wind field and a time-varying wind field obtained from the numerical weather prediction model INCA. The results show significant differences in the deposition patterns and snow depth even after a 6 hour drift period. Using time-averaged boundary conditions leads to an underprediction of the resulting snow depth caused by averaging the wind speed, which lets gusts of wind disappear while snow transport is a non-linear function of the wind speed. Using numerical weather prediction models for snowdrift simulation therefore provides enhanced knowledge of the snow depth for local avalanche warning services.

SnowDrift3D, a high-resolution, atmospheric snow-transport model, is presented for the first time. In contrast to most state-of-the-art snowdrift models, atmospheric particle transport, i.e. saltation and suspension, is accounted for by one passive transport equation. The model uses unsteady wind fields (spatial resolution of up to 2 m) computed with an atmospheric computational fluid dynamics model that is directly connected to the numerical weather prediction model ALADIN. Sensitivity runs show that (1) the saltation mass flux is a function of cubic shear velocity, , (2) the model is marginally sensitive to the grid spacing at high resolutions (up to 2 m), (3) the model computes the redistribution of snow at high resolution in real time on dual core personal computers and (4) the changing topography of the snow cover should be included in cases of local erosion or deposition of a large amount of snow. Finally, we present a comparison of modeled and measured snow distributions obtained by terrestrial laser scanning showing area-wide linear correlation up to R = 0.33.