The project started in 2019 and was born from an initiative of the director of the SLF, Jürg Schweizer. A team of SLF researchers and avalanche forecasters worked on it, together with colleagues from the Swiss Data Science Center. For two years, physicist Cristina Pérez experimented with different methods and datasets, processed the data and finally trained the model with it. To do this, it used meteorological data and snow cover simulations dating back twenty years, based on measurements from the intercantonal measurement and information system IMIS. This approach is called Machine Learning. Among the challenges, on the one hand it was necessary to choose the parameters so that the algorithms were more and more precise. “On the other hand, it was difficult to obtain good accuracy for avalanche warning level four, because this high alert level has appeared only rarely over the last twenty years, the database was therefore quite reduced,” explains Pérez. The collaborators of the avalanche forecasting service call Palantir the platform on which they consult the various ML models, based on the seven crystal balls of the fantasy world Arda by JRR Tolkien, whose best-known continent is Middle-earth, and which show scenes very distant in space and time.
Of course, the human collaborators of the avalanche forecasting service use the same data and models as the computer for their work. But they additionally use information such as current field observations and feedback. The computer does not have this data. The algorithm is based exclusively on snow cover simulations as input. On the other hand, if only for reasons of time, men choose from the quantity of data those which are relevant for them, the machine does not make a selection. “The models enable a spatial and temporal resolution that we humans will never achieve,” says Techel. Man and machine complement each other. Algorithms help interpret basic data sets. Both sides also make mistakes. “The good news is that models make different mistakes than we do,” says Techel. The avalanche forecast service thus obtains a second independent opinion and can reconsider its current result for the avalanche bulletin in the event of significant discrepancies.
The team is currently continuing to develop the project and wants to better combine human and mechanical predictions in the future. “This also means a more intuitive presentation of results for the avalanche forecast service,” explains Techel.
