The Application of Multi-Temporal Unmanned Aerial Vehicle (UAV) Imagery for Automated Landslide Identification

Abstract

As extreme rainfall events driven by climate change become more frequent and severe, the window for effective disaster response narrows. Timely post-event assessments of natural hillside conditions are therefore increasingly critical for effective landslide risk management. Unmanned Aerial Vehicle (“UAV”)-based surveys coupled with an automatic landslide detection algorithm offer high-resolution and accurate terrain data necessary for rapid landslide identification. However, the strict limitation of UAV operation of flight height and Visual Line of Sight (“VLOS”) hinders efficient data acquisition in remote terrains or suburban areas. Under the Low-altitude Economy Regulatory Sandbox (“Sandbox”), it permits pilot trials for the unconventional UAV Beyond Visual Line of Sight (“BVLOS”) operations, including extended flight height to 1000ft and automated UAV flights control by an off-site computer control platform. These break-through applications open up the opportunity for using UAVs for reliable, quick and efficient large area landslide identification. The paper presents the findings of the study, which leverages the Sandbox application for BVLOS operation. Based on the data obtained from pre- and post-rainstorm surveys by UAV, an automated algorithm was developed by comparing the Normalised Difference Vegetation Index (“NDVI”) georeferenced images and the high-density photogrammetry point clouds to identify terrain surface changes for the formulation of a fully automated landslide identification workflow.