Deciphering Landslide Precursors From Spatiotemporal Ground Motion Using Persistent Homology

Abstract

Landslides are major natural disasters that pose significant challenges for prediction. Recent advances in monitoring tools have led to the accumulation of monitoring data with high spatiotemporal resolution, calling for new and robust methodologies to efficiently analyze these complex big data and accurately predict landslides. Here, we present a persistent homology-based method that integrates the slope-scale monitoring data from interferometric synthetic aperture radar with novel measures of spatiotemporal evolution of slope deformation to identify early warning precursors for impending landslides. Our proposed method can capture critical patterns of accelerated deformation evolution and generate warning signals long before the landslide occurrence. Six case studies confirm the effectiveness and accuracy of the proposed method in landslide prediction, with a leading time exceeding 100 days for the Xinmo and Mud Creek landslides. Strong spatiotemporal correlations of slope deformation underscore long-range effective predictions. Our method offers a new, robust alternative to the conventional threshold-based approach for understanding and predicting landslides in natural slopes.