Improving the spatial distribution of snow height in physics-based snow models using large-area airborne lidar-scans

Predicting the volume and timing of snowmelt is essential for applications such as hydropower production planning and flood forecasting. The timing of snowmelt is strongly influenced by the spatial distribution of snow. A more heterogeneously distributed snowpack leads to a longer melt season and lower peak flow than a homogeneously distributed snowpack. Despite the importance of spatial snow distribution for runoff characteristics, large-scale and high-resolution measurements of snow distribution are rare and it is challenging to effectively use such measurements in models when the snow conditions differ substantially across the study region.

We acquired high-resolution airborne lidar snow height maps in three winters for three large hydropower regions in Southern Norway, covering over 1000 km². We use these to improve snow height simulations and demonstrate how the scans can be assimilated into a physics-based snow model. To this end, we use a snowfall scaling method that aims to implicitly describe preferential deposition and redistribution processes during snow accumulation by altering the snowfall inputs to the snow model. In each grid cell, a scaling factor is chosen such that the modelled snow height matches the observed snow height. Existing methods are often not finding the optimal scaling factor, especially in case snowmelt has started in parts of the scanned regions. We present a new approach that considers estimated snow losses due to melt and sublimation that occurred before the acquisition of the lidar scan. With this improvement, scans taken slightly after melt onset in part of the region can still be used to reliably find the optimal snowfall scaling factors, even if part of the snow is already lost due to melt and sublimation.

We show how similar these snowfall scaling factors are between years, due to repeatable patterns in snow height, and whether this similarity provides opportunities to transfer snowfall scaling factors to different years. Furthermore, we show that higher model resolutions are best suited to represent the observed spatial snow distribution in the model using the proposed snowfall scaling method. The insights of this work can be used to effectively use large area, high-resolution snow height measurements in snow models.

This work is partially funded by Statkraft Energi AS and the Norwegian Research Council (SnowInflow, NFR 346308).