High-precision photogrammetry for glaciology
PhD: University of Colorado Boulder, 2018.
Consumer-grade digital cameras have become ubiquitous tools for documenting short-term variability in the geosciences. However, these devices were not intended for precise timekeeping and surveying, and their use as such requires management of systematic and random errors that inevitably arise.
This dissertation presents a suite of methods for registering the place and time of photographs in an absolute reference frame so that they may be analyzed and interpreted alongside other spatial and temporal data. The methods are tested on a 13-year record of 33,000 time-lapse photographs from Alaska’s Columbia Glacier. This work provides insights into the capabilities and shortcomings of consumer-grade cameras as scientific instruments, the opportunistic approaches often needed to achieve the best results, and the potential of continuous high-frequency measurements for documenting rapid geomorphic processes.
Subsecond-precision image capture times are achieved by measuring the offset to a reference clock display and accounting for the drift, precision, and reporting resolution of the camera clock. Two case studies illustrate the benefit of subsecond precision in contemporary investigations: georeferencing aerial photogrammetric surveys with camera positions time-interpolated from GPS tracklogs, and coupling videos of glacier-calving events to synchronous seismic waveforms. Retroactive dating of photographs, on the order of seconds to hours, is achieved by leveraging phenomena visible in the photographs – namely, the positions of astronomical objects in the sky or the corresponding variations in solar radiation and sea level. Similarly, retroactive camera calibrations are achieved using surface and topographic features in the photographs – specifically, point and line features of known absolute position, the motion of static features in images due to camera rotation, and the correspondences between real images and images synthesized from vertical imagery. Camera motion is corrected by computing globally optimal estimates of rotation over arbitrarily-long photographic sequences.
Finally, a recently-developed tracking algorithm based on particle filtering theory is refined and applied to estimate Columbia Glacier velocities, their associated uncertainties, and the corresponding strain rate fields at 3 d intervals over a 13-year period, providing an unprecedented look at the seasonal and sub-seasonal variability of tidewater glacier dynamics over long time scales.