EXPLORING BEYOND 3D DATA FOR THE AUTOMATED DETECTION OF LANDSLIDES USING PERSISTENT HOMOLOGY

Landslides are major natural disasters that can result in a considerable amount of damage to property, infrastructure, residential structures, and ecosystems, can block transportation routes, and in worst cases result in the loss of human lives. Landslide inventories provide spatiotemporal information about historical and more recent landslides and are used for analysis to create susceptibility models that are an essential resource for risk management tasks. Most landslide inventories are created by detecting changes over time, which requires multiple datasets that may be hard to acquire, or manual detection methods, which require considerable amounts of time, money, and labor to complete. A semi-automated/automated detection algorithm would significantly increase efficiency, reduce costs, and decrease labor needed to compile a landslide inventory. In our prior work we developed a method that utilized a topological tool - Persistent Homology – to automatically detect landslides by using points gathered from a ridgeline detection algorithm applied to Light Detection and Ranging (LiDAR) elevation data. Persistent Homology is a tool for recording topological features and reducing the features of a dataset to their most simple forms and providing information on the length of feature persistence, which provides insights into their size and type. In the case of landslide detection, small topographic anomalies automatically detected from LiDAR point elevation data can also highlight characteristic perimeter features of a landslide (scarp, flank, and toe). From the resulting points - that preserve three dimensions (spatial location and elevation) - the application of the Persistent Homology method allowed for the identification of patterns that are likely associated with landslides, which in this case would be three dimensional ‘holes’ generated by the perimeter of a landslide. What has not been explored yet is: Would the use of additional dimensions, beyond the spatial and elevation inherent in topographic datasets, impact the accuracy of detecting landslides using Persistent Homology? In this work, we report on the result of this exploration for a region in Northeastern Pennsylvania.