Generating maps for operational avalanche warning with machine learning algorithms

184-188

Artikel aus einem Buch

200213011

Snow avalanche warning services use a wide range of tools to ensure that users of their bulletins receive reliable and consistent information. This study describes a new tool for predicting maps of danger levels, avalanche problems, snowpack stability, and trends for incidents for the upcoming 2.5 days. The input consists of topographical data, meteorological data from the nowcast tool INCA, and data from the numerical weather prediction model AROME. Additionally, snow depth data available on a 1 km grid from SNOWGRID-CL is used. Machine learning methods were applied. The most suitable algorithm was determined for each target variable (danger level, avalanche problems, extended column tests, incident trend). The results are detailed maps for the development of the target variables for the next 2.5 days. During the model development, preliminary results were shared with practitioners, so that feedback could be utilized immediately.
KEYWORDS: Prediction, risk assessment, random forest, support vector machine, nearest neighbor, multi-output classifier