Estimating three-dimensional motion of a creeping landslide from topographic data and associated land surface parameters

Abstract

Recent advances in surveying technologies have allowed high precision measurement and monitoring of changes in the Earth's surface position over time. Lateral mass movements remain an under-explored aspect of topographic analyses despite the plethora of dynamic processes affecting surface position. This research introduces the use of a two-dimensional optical flow algorithm to estimate the three-dimensional relationships between a topographic surface and itself after displacement using a time series of bare-earth digital elevation models (DEMs). Several indices are derived from the motion fields estimated by the optical flow algorithm to define a set of properties that are used to quantitatively characterize surface motion. A preliminary investigation into the efficacy of these surface properties for analysing dynamic topography was conducted on a creeping landslide in Biratori, Hokkaido, Japan. An accuracy assessment demonstrated strong agreement between observed and estimated displacements, with concordance correlation coefficients of 0.87 for both $x$- and $y$-axis displacement, and submetre root mean squared error of 0.47 and 0.72 m for the $x$- and $y$-axis, respectively. A segmentation algorithm was applied to the translation distance and azimuth angle properties to assess the accuracy with which these variables delineate the landslide, based on the expectation that landslide motion characteristics are spatially contiguous and internally homogeneous. Segments overlapped with the landslide boundary area by up to 70%, and segments within the landslide boundary were consistently among the largest in the segment-area distribution. The results demonstrated how these surface properties can form simple and effective heuristics for analysing creeping landslides with strong potential for other dynamic surface phenomena.