Using machine learning to estimate bulk snow density based on weather data
Measurements of bulk snow density are costly but required for converting snow depth to snow water equivalent, the relevant quantity for hydrologists and water resource managers. The bulk density of snow develops over the winter season by gravitational settling but also in response to weather conditions, for instance by wind hardening or ice layer formation due to rain-on-snow events.
Here we aim for a simple parameterization of snow density as a function of weather variables, using machine learning algorithms to select those variables having the highest predictive power.
The results will be checked against those from the detailed snowpack model CROCUS and available field data. Such a simplified parameterization may prove very useful in a range of applications where the mass of snow has to be determined from snow depth measurements (e.g. snow hydrology, glacier mass balance).