Statistical classification of self-organized snow surfaces

Wind-swept snow forms organized textures called bedforms. These bedforms ornament about 8% of the surface of the Earth. They alter the thermodynamic properties of the surface, and affect local and global energy fluxes. Their effects have not yet been incorporated into Earth system models because the conditions governing their development are not yet well understood.

We present the first rules predicting the formation, or not, of snow bedforms in a range of weather conditions. These rules are generated from three years of observations of snow bedforms in the Colorado Front Range. We identified the weather conditions leading to each observation, and used this data to construct classifiers, validated by bootstrapping, that predict bedform growth as a function of weather. Finally, we identified the variables which best predict bedform presence by constructing and comparing 460 classifiers built from different combinations of weather variables.

We find that bedform presence is well predicted by wind speed and time since snowfall. The presence of sastrugi, the most widespread and studied bedform, is best predicted by the time and the highest wind speeds since snowfall.

Our results make it possible to forecast bedform extents from widely-available weather data. They represent a first step towards understanding a process that shapes much of the polar world.