Seismic precursors to the Blatten, Switzerland landslide revealed by unsupervised machine learning

The transition from stable to unstable states in geological systems, such as landslides and fault zones, remains poorly understood. Seismic precursors and foreshocks related to the transition are often difficult to observe and the interpretation remains challenging. Here, we report an observation of the nucleation process preceding the glacial landslide on May~28,~2025 in the village of Blatten, Switzerland. We identify three phases using an unsupervised machine learning approach applied to 20 days of continuous seismic data recorded before the main event. We separate the rockfalls from the seismic signature associated with glacier sliding. We interpret it as a slip-weakening behavior and acceleration in slip during the last two days ahead of the glacial failure. These results demonstrate the potential of unsupervised learning to classify such seismic precursors in advance of the collapse, offering promising implications for early warning systems and landslide risk mitigation.