Artificial intelligence supported extended range Doppler radar: avalanche activity measurement and mitigation verification in Evolène, Switzerland

The municipality of Evolène, located in the Swiss Alps, is exposed to significant avalanche hazard. Evolène comprises several settlements situated at the base of a mountain slope approximately 4.5 km in length with a topographic relief of 2000 m. During the winter of 1999, an intense storm cycle with significant snowfall accumulation rates triggered a destructive avalanche cycle with multiple fatalities and significant damage to property and supporting infrastructure. The slope affecting the municipality is characterized by numerous small and spatially distributed avalanche start zones, resulting in multiple avalanche paths that pose a significant and persistent threat to the valley’s population and infrastructure networks. As a result of the 1999 event, defense measures have been implemented to contribute to the overall mitigation scheme dealing with the avalanche hazard. These measures have been implemented over the past two decades, where a comprehensive avalanche protection and risk mitigation program has been deployed to reduce residual risk within the municipality. In the early 2000s, extensive structural measures, primarily snow-supporting/retaining measures in the start zones, were rapidly deployed. In a more recent project phase, the protection concept was expanded to include operational measures, notably remote avalanche control systems (RACS) and radar-based avalanche detection. In 2023, five RACS units and two radar systems were installed, with approximately twenty additional RACS units planned for phased installation in the coming years. The two radar systems are horizontally combined to provide continuous coverage of the entire slope, enabling the detection of avalanche events under all meteorological conditions at ranges exceeding 5.5 km. To enhance detection performance beyond conventional signal-processing algorithms, Geoprevent integrated a convolutional neural network–based artificial intelligence model, improving system sensitivity and reducing false detections. The system is actively used by local avalanche forecasters and practitioners for operational avalanche control and decision support, and delivered critical observational data during the active avalanche period of mid-April 2025. Although further improvements in detection accuracy remain possible, the generated data constitutes a valuable basis for evaluating system performance and for validating the spatial placement of current and planned RACS installations.