Forecasting landslides using community detection on geophysical satellite data.

Area of Science:

• Geophysics
• Network Science
• Geological Hazard Assessment

Background:

• Natural hillslopes are susceptible to dramatic failure due to extreme weather events like heavy precipitation.
• Predicting the transition from gradual ground deformation (creep) to sudden, catastrophic slope failure remains a significant challenge.
• Prefailure deformation patterns are often only apparent in retrospect, hindering timely risk assessment.

Purpose of the Study:

• To investigate spatiotemporal deformation patterns preceding landslides using a network science approach.
• To develop a predictive metric for catastrophic slope failure.
• To identify and assess the risk of both failed and creeping landslides.

Main Methods:

• Transforming satellite radar data (InSAR) and digital elevation model data into a spatially embedded network.
• Utilizing multilayer modularity optimization to identify strongly connected communities (clusters) of ground deformation.
• Weighting network edges by local slope and ground surface deformation to represent failure susceptibility and rheological state.

Main Results:

• Successfully identified the location of the Mud Creek landslide and nearby creeping landslides.
• Developed a 'community persistence' metric to quantify deformation patterns leading up to failure.
• Observed a significant increase in community persistence in the weeks preceding the Mud Creek landslide.

Conclusions:

• Network science methods, specifically multilayer modularity optimization, show promise for landslide prediction.
• The community persistence metric can quantify pre-failure ground deformation patterns.
• This approach offers a novel technique for highlighting regions at high risk of catastrophic slope failure.