Evaluation of probabilistic snow avalanche simulation ensembles with Doppler radar observations
Journal article
Permanent lenke
https://hdl.handle.net/11250/3087443Utgivelsesdato
2013Metadata
Vis full innførselSamlinger
- NGI articles [1085]
Originalversjon
Cold Regions Science and Technology 97(2014), 151–158. doi:10.1016/j.coldregions.2013.09.011Sammendrag
In recent years snow avalanche simulation software, which is based on depth-averaged models operating in three-dimensional terrain, has gained importance. Simulations are used to plan protection measures. The software computes the spatio-temporal evolution of flow depth and velocity and is optimized to determine runout distances of avalanches. However, considering the complex output of these computer models no sophisticated evaluation procedure exists to compare the velocity output with experimental data. In this study we present a new method to objectively evaluate the velocity results of the simulation software SamosAT with Doppler radar measurements from two European snow avalanche test sites. A coordinate transformation allows comparing the maximum simulated velocities with measured data. A probabilistic simulation approach is applied, performing and evaluating a large number of simulation runs. The simulation results are evaluated in two ways. Firstly, an average evaluation is used to obtain a discrepancy estimate, representing the deviance between simulations and measurements and to explore the range of possible velocity results accounting for different uncertainties in the input parameters. Secondly, an optimized evaluation determines the magnitude of possible accordance between simulation and measurement. It reveals that for the employed simulation approach release depth and certain friction coefficients are crucial parameters to obtain optimal correspondence of simulated and measured velocities. In the investigated cases, simulation and observation show a reasonable accordance with a tendency of velocity underestimation. The presented approach is a valuable method for the evaluation and comparison of complex model outputs.