Comparison between high-resolution large-eddy simulation (LES) of atmospheric flows and the WindNinja wind downscaling algorithm, emphasizing Lagrangian coherent structures to enhance predictions for snow redistribution in alpine environments.

Abstract

Avalanche forecasters, hydrologists and winter recreationalists are asking themselves the same question: where is the snow on the mountain. Unfortunately, snow distribution is a highly complex area of research due to the turbulent behavior of atmospheric flow. The structural organization of the flow can be analyzed via Lagragian coherent structure (LCS) methods, such as finite-time Lyapunov exponents (FTLE), and complemented by Eulerian diagnostics like the maximal eigenvalues of the rate-of-strain (ROS) tensor. Nevertheless, such analyses are only as reliable as the underlying wind velocity field they are based on.

The diagnostic wind model WindNinja was developed as a wind downscaling tool to estimate flows in mountainous regions from low resolution forcings, such as from a numerical weather forecast. However, performance limitations become evident, especially when compared to higher-fidelity numerical models that more accurately resolve nonlinear aspects of the flow, such as Large-eddy simulations. Previous work has shown the benefit of informing WindNinja downscaling with local point initialization data in addition to the bulk forcing, but no systematic best practice has been developed. The present thesis seeks to improve WindNinja's accuracy by providing point initialization data in regions of significant Lagrangian coherent structures. Local "weather station" positions are selected based on the position of FTLE or ROS-eigenvalue ridges identified in reference Large-eddy simulations. Local forcing data for the weather stations is provided from the Large-eddy simulation as well.

Efforts show that strategically placed weather stations could improve overall wind speed and slightly reorient flow channels, but were unable to generate flow recirculation cells. Furthermore, WindNinja’s strong dependence on the terrain and its limitations as a steady-state model, which leads to a complete neglect of wind sheltering and blockage, is demonstrated.

It can be concluded that WindNinja may be suitable as a rough suggestion of snow accumulation regions but can not faithfully recreate conditions for accurate quantification of accumulation rates or flux directions.