Towards statistical alarm threshold determination for alpine rock fall monitoring systems

Alpine rock slope failures are frequent hazardous events. To reduce risk, unstable rock sections are increasingly monitored, e.g. with automatic displacement and tilt meters or high-resolution remote techniques. However, effective mitigation actions require the triggering of meaningful alarms. Currently, these alarm thresholds are often set manually based on expert knowledge, which may create too conservative or insufficiently sensitive thresholds that are not well enough adapted to changing conditions over time. Instead, we hypothesise that dynamically updated thresholds based on statistical analyses of continuous observations can provide a more robust, comprehensible, and performant approach to early warning.

Here, we present an approach to determine alarm thresholds for automatic monitoring devices based on statistical analyses of past observation data of two high-alpine sites. We analyse multiple years of automatic measurements gathered from high-frequency real-time monitoring systems. At the Hochvogel summit (DE/AT; 2,592 m a.s.l.), we monitor a 200,000 to 600,000 m³ complex rock slope instability with 12 sensors since 2019 and no major rock fall event yet. The second site at the Kitzsteinhorn north flank (AT; 3,029 m a.s.l), includes 6 sensors and a 600 m³ rockslide failure in August 2025 that has been recorded by the displacement sensors. Preliminary results show that this approach is able to reduce the frequency of false alarms over time and can detect critical accelerations earlier than fixed manual thresholds.

We anticipate that this statistical analysis of multiple years of observations including failure and non-failure events can guide decision-makers and monitoring system operators on how to set initial reasonable alarm thresholds and how the thresholds can be adjusted over months and years of system operation for an early detection of hazardous accelerations.