A deep artificial neural network has been developed to calculate snow profiles, avalanche danger levels, avalanche problems, and indices for snow avalanche occurrences. This network uses topographic data from a digital elevation model (10x10m) and meteorological recordings from automated weather stations as input. The output variables in the snow profiles are snow temperature, hardness, grain shape, and grain size. The artificial neural network was trained with 3342 snow pit observations from 2016 to 2021 in Austria (Tyrol, Salzburg). The surrounding area (100x100m) of a snow pit observation is considered and treated by convolutional layers to extract topographic features. For each meteorological variable (air temperature, humidity, wind speed, wind direction, snow depth) the three closest stations were used with their horizontal and vertical distances as additional input. These time series are processed by recurrent layers to consider their temporal evolutions (672 hours backwards). For each snow pit observation site, the corresponding avalanche danger level, the avalanche problems, and indices obtained from recordings of the avalanche warning services and avalanche accidents (Austrian Board for Alpine Safety) are used.
