Comprehensive exploitation of seismic data at a debris flow torrent – Early warning, hazard assessment, parameter estimation

Abstract:

Debris flows are fast-moving mixtures of water, mud, and fragmented rock traveling down steep torrents. Triggered by heavy precipitation, they constitute a risk to human life and property. Fast detection of debris flows and quick assessment of their destructive potential are crucial for early warning, evacuation plans, and damage mitigation. In recent years, seismology has evolved into a non-invasive tool, which does not require challenging in-torrent installations and may be an alternative to, e.g. costly radar and ultrasonic flow stage measurements. Here, we make use of a seismic data set of more than 20 debris flows recorded with an eight-station network at the Illgraben torrent in Switzerland. We use debris flow records from four summers to show that such a local seismic network can a) increase early warning time, b) elucidate erosional patterns during the flow and c) give an estimate of debris flow parameters, such as particle sizes and flow stage.

For early detection of debris flows, we use the random forest algorithm to classify windowed seismic data on the continuous real-time data stream. This method increases early warning time by tens of minutes compared to traditional in-torrent early warning systems. We investigate debris flow dynamics from the high-frequency seismic signature by inverting for vertical and horizontal forces generated by particle impacts and shearing. The temporal evolution of the inverted forces constrains deposition patterns and particle sorting along the channel. Finally, tilt signals from a broadband seismometer close to the channel provides additional information on debris-flow volume and density. The seismic data thus does not only facilitate detection, but constrains parameters that identify particularly destructive events.