Unmanned Aerial Vehicle Remote Sensing of Shallow Snow: Assessment and Possibilities for Improved Snow Depletion Prediction

Abstract:

Unmanned Aerial Vehicles (UAVs) have been enthusiastically adopted by many earth scientists due to their ability to provide Digital Surface Models (DSM) and orthomosaics of unprecedented spatial and temporal resolution. These datasets have great potential to advance the prediction of snow hydrology in particular but have had little testing in areas of shallow snowcover. To assess the utility and possibilities of UAV data products for quantifying and predicting the properties and processes of shallow snowcovers, an intensive field campaign took place during the 2015 melt season in a prairie agricultural field in Saskatchewan, Canada. The wheat field with standing stubble (15-35cm) had little topographic relief, a shallow snow (peak <40cm) and became patchy as snowcovered area declined during melt. Over the 25-day melt period 24 flights were performed with a Sensefly Ebee UAV to map the 120 hectare area. Structure from motion techniques, as implemented in Postflight Terra 3D software, generated DSMs and orthomosaics at a 3.5 cm resolution. Orthomosaic analysis quantified snowcovered area at unprecedented accuracy and frequency allowing for new insights into the spatial characteristics of the snowcover depletion process. However, vertical errors of the DSMs were significant (root mean square error of 10-20cm) compared to snow depth, making any comparisons of DSMs too uncertain to be useful for estimating ablation directly. In contrast, the relative accuracy of the DSMs was sufficient to calculate a DSM roughness index that relates to the coefficient of variation of snow water equivalent (SWE), as measured from intensive ground measurements. The DSM roughness index alone can predict the shape of the snow depletion curve and, if used in conjunction with point measured or modelled SWE, can calculate the SWE frequency distribution and snowcover depletion. It is proposed that UAV derived surface roughness, in conjunction with point modelled or observed SWE can provide a simple method to predict areal snowcover depletion without undertaking extensive manual snow surveying, even in shallow snowpack environments where direct measurement of snow surface ablation from UAVs is not possible.