Cordilleran Section - 115th Annual Meeting - 2019

Paper No. 16-5
Presentation Time: 9:25 AM

NEW DATA FROM OLD DATA: ADVANCING 3D CHANGE DETECTION FROM ARCHIVAL AERIAL PHOTOGRAPHS THROUGH A MODIFIED STRUCTURE-FROM-MOTION IMPLEMENTATION


BEMIS, Sean P.1, VADMAN, Michael J.2 and MELSEN, Christopher2, (1)Global Forum on Urban & Regional Resilience, Virginia Tech, Blacksburg, VA 24060, (2)Department of Geosciences, Virginia Tech, Blacksburg, VA 24060

Inspired by Ray Weldon’s vision for extracting valuable crustal deformation data from archival datasets, we explore new avenues for robust topographic/geomorphic change detection from archival aerial photographs. Advances in photo-based 3D reconstruction techniques, such as Structure-from-Motion + MultiView Stereo (SfM+MVS), simplify the creation of historical orthoimagery and digital topography, but small errors in scaling make change detection using these datasets problematic. We adopt a modified implementation of SfM processing procedures, referred to here as Time-SIFT, to improve topographic change detection using archival aerial photographs by simultaneously aligning multiple generations (epochs) of photographs as a single processing block. Once aligned, we generate dense point clouds for each epoch individually. These point clouds, or their derivatives, are then differenced to measure volumetric change. We test the Time-SIFT method by applying it to several locations where we acquired sequences of aerial photograph datasets and recent high-resolution topography that span >50 years. Our Time-SIFT processing of 1955, 1970, 1979, and 1983 aerial photographs and 2007/08 lidar of Mount Rainier reproduce glacial ice volume changes determined by Sisson et al. (2011) for 1970 to 2007/08 and provide an expanded time series of the ice volume change. Our initial attempts to measure the volume of edifice change associated with the 1980 eruption of Mount St. Helens illustrate a key limitation of the Time-SIFT procedure, where excessive changes across the landscape prevent the pre- and post-eruption photographs from aligning simultaneously. We also found that surface translations, such as translational landslides, some modes of glacier motion, and horizontal tectonic displacements, create another challenge for Time-SIFT studies because of how photogrammetry derives elevation from parallax. These surface translations can introduce a false parallax, potentially resulting in regions of erroneous elevations. Despite these challenges, Time-SIFT methodology has the potential to extend change detection analyses for some geomorphic and tectonic processes by several decades, being limited primarily by the availability and scale of archived photographs.