Rock slope failure preparation paced by total crack boundary length

Gravitational mass wasting prediction requires understanding of the factors controlling failure. Prior to slope failure, cracks in weakened rock are thought to grow and coalesce, eventually forming a continuous failure plane. Here, a hidden Markov machine-learning model is applied to seismic data to reveal the temporal evolution of cracks prior to a major rockslide event in the Swiss Alps. After a prolonged linear increase of the crack cumulative number, an S-shaped crack rate pattern occurred in the day before the rockslide. A simple mechanistic model explains this behaviour, showing that total crack boundary length is the key factor controlling failure plane evolution immediately before mass movement. The findings imply that cracks should be treated as 2-D, rather than 1-D objects, and that slope failure can be driven predominantly by internal rather than external processes. The model offers a novel, physically based approach for early warning of slope failures.