Debris flow warning system (SALT)

The aim of the Système d’Alerte de Laves Torrentielles (SALT) project is to set up a meteorological warning system for selected sites throughout the canton of Valais. These areas have been affected by numerous debris flow events in the past, which is why a real-time warning system project has been initiated to prevent such events in the future. This system enables SMS and/or email alerts to be sent to a defined list of people when a precipitation intensity threshold has been reached. If necessary, these alerts can automatically trigger road closure systems, for example.

In the first phase (SALT-I), 26 sites were put into operation, with a similar precipitation intensity threshold for all zones. The performance of the warning system was calculated for the test phase in summer 2023. This showed that the warning system predicted debris flows correctly at around 35%, and that 65% of events were missed. The system also issued some false alarms. These results show that the application of a single intensity threshold is not optimal. They can be explained by the fact that debris flows take place in sites with different geomorphological characteristics, and that variables other than rainfall can have an impact on their triggering.

In order to enhance the performance of the warning system, a second phase of the project (SALT-II) aims to define regional intensity thresholds for the entire canton of Valais. To do this, we will use the debris flow entries in the canton’s event register, as well as historical precipitation data.

This project also aims to develop a Machine Learning model with the aim of developing a system capable of integrating local conditions and other meteorological parameters, such as previous rainfall, presence or absence of snow, temperature, soil saturation, etc. This model could predict debris flows even more accurately than the method based on precipitation thresholds. This model could predict debris flows even more accurately than the method based on precipitation thresholds.